Merge remote-tracking branch 'origin/litellm_internal_staging' into litellm_/elegant-edison-bf44a2

.github/actions/detect-backend-changes/action.yml

@@ -0,0 +1,48 @@
+name: "Detect backend-relevant changes"
+description: >-
+  Classify the pull request's changed files with .circleci/scripts/classify_changes.sh
+  and expose decision=run|skip. decision=skip means only ui/**, **.md or **.mdx files
+  changed, so callers can short-circuit expensive steps while the job still completes
+  successfully and satisfies its required status check. The decision defaults to run for
+  any non pull_request event or whenever the changed set cannot be resolved, so tests are
+  never skipped when the classification is uncertain.
+
+outputs:
+  decision:
+    description: "run when backend-relevant files changed, otherwise skip"
+    value: ${{ steps.classify.outputs.decision }}
+
+runs:
+  using: composite
+  steps:
+    - id: classify
+      shell: bash
+      env:
+        BASE_SHA: ${{ github.event.pull_request.base.sha }}
+      run: |
+        set -uo pipefail
+        if [ -z "${BASE_SHA:-}" ]; then
+          echo "detect-backend-changes: not a pull_request event; running job"
+          echo "decision=run" >> "${GITHUB_OUTPUT}"
+          exit 0
+        fi
+        if ! git fetch --no-tags --depth=1 origin "${BASE_SHA}" >/dev/null 2>&1; then
+          echo "detect-backend-changes: could not fetch base ${BASE_SHA}; running job"
+          echo "decision=run" >> "${GITHUB_OUTPUT}"
+          exit 0
+        fi
+        changed="$(git diff --name-only "${BASE_SHA}" HEAD 2>/dev/null)" || {
+          echo "detect-backend-changes: git diff failed; running job"
+          echo "decision=run" >> "${GITHUB_OUTPUT}"
+          exit 0
+        }
+        if [ -z "${changed}" ]; then
+          echo "detect-backend-changes: no changed files vs ${BASE_SHA}; skipping job"
+          echo "decision=skip" >> "${GITHUB_OUTPUT}"
+          exit 0
+        fi
+        echo "detect-backend-changes: changed files vs ${BASE_SHA}:"
+        printf '%s\n' "${changed}" | sed 's/^/  /'
+        decision="$(printf '%s\n' "${changed}" | bash .circleci/scripts/classify_changes.sh backend)" || decision="run"
+        echo "detect-backend-changes: decision=${decision}"
+        echo "decision=${decision}" >> "${GITHUB_OUTPUT}"

.github/workflows/_test-unit-base.yml

@@ -45,12 +45,18 @@ jobs:
     name: Run tests
     runs-on: ubuntu-latest
     timeout-minutes: ${{ inputs.timeout-minutes }}
+    outputs:
+      decision: ${{ steps.changes.outputs.decision }}
 
     steps:
       - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
         with:
           persist-credentials: false
 
+      - name: Detect backend-relevant changes
+        id: changes
+        uses: ./.github/actions/detect-backend-changes
+
       - name: Set up Python
         uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
         with:
@@ -72,16 +78,19 @@ jobs:
             ${{ runner.os }}-uv-
 
       - name: Install dependencies
+        if: steps.changes.outputs.decision != 'skip'
         run: |
           .github/scripts/uv_sync_with_retries.sh --frozen --group ci --group proxy-dev --extra google --extra proxy --extra semantic-router
 
       - name: Generate Prisma client
+        if: steps.changes.outputs.decision != 'skip'
         env:
           PRISMA_BINARY_CACHE_DIR: ${{ runner.temp }}/prisma-cache
         run: |
           uv run --no-sync prisma generate --schema litellm/proxy/schema.prisma
 
       - name: Run tests
+        if: steps.changes.outputs.decision != 'skip'
         env:
           TEST_PATH: ${{ inputs.test-path }}
           MAX_FAILURES: ${{ inputs.max-failures }}
@@ -114,7 +123,7 @@ jobs:
           fi
 
       - name: Save coverage report
-        if: always()
+        if: always() && steps.changes.outputs.decision != 'skip'
         uses: actions/upload-artifact@4cec3d8aa04e39d1a68397de0c4cd6fb9dce8ec1 # v4.6.1
         with:
           name: coverage-${{ inputs.artifact-name }}-${{ github.run_id }}-${{ github.run_attempt }}
@@ -124,7 +133,7 @@ jobs:
   upload-coverage:
     name: Upload coverage to Codecov
     needs: run
-    if: always()
+    if: always() && needs.run.outputs.decision != 'skip'
     runs-on: ubuntu-latest
     permissions:
       contents: read

.github/workflows/test-unit-documentation.yml

@@ -32,6 +32,10 @@ jobs:
           path: docs/my-website
           persist-credentials: false
 
+      - name: Detect backend-relevant changes
+        id: changes
+        uses: ./.github/actions/detect-backend-changes
+
       - name: Set up Python
         uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
         with:
@@ -53,10 +57,12 @@ jobs:
             ${{ runner.os }}-uv-
 
       - name: Install dependencies
+        if: steps.changes.outputs.decision != 'skip'
         run: |
           .github/scripts/uv_sync_with_retries.sh --frozen --group ci --group proxy-dev --extra google --extra proxy --extra semantic-router
 
       - name: Generate Prisma client
+        if: steps.changes.outputs.decision != 'skip'
         env:
           PRISMA_BINARY_CACHE_DIR: ${{ runner.temp }}/prisma-cache
         run: |
@@ -64,6 +70,7 @@ jobs:
 
       # Run the same documentation tests that CircleCI ran (as direct Python scripts)
       - name: Run documentation validation tests
+        if: steps.changes.outputs.decision != 'skip'
         run: |
           uv run --no-sync python ./tests/documentation_tests/test_env_keys.py
           uv run --no-sync python ./tests/documentation_tests/test_router_settings.py

.github/workflows/test-unit-proxy-legacy.yml

@@ -49,6 +49,10 @@ jobs:
         with:
           persist-credentials: false
 
+      - name: Detect backend-relevant changes
+        id: changes
+        uses: ./.github/actions/detect-backend-changes
+
       - name: Set up Python
         uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
         with:
@@ -70,16 +74,19 @@ jobs:
             ${{ runner.os }}-uv-
 
       - name: Install dependencies
+        if: steps.changes.outputs.decision != 'skip'
         run: |
           .github/scripts/uv_sync_with_retries.sh --frozen --group ci --group proxy-dev --extra google --extra proxy --extra semantic-router
 
       - name: Generate Prisma client
+        if: steps.changes.outputs.decision != 'skip'
         env:
           PRISMA_BINARY_CACHE_DIR: ${{ runner.temp }}/prisma-cache
         run: |
           uv run --no-sync prisma generate --schema litellm/proxy/schema.prisma
 
       - name: Run tests - ${{ matrix.test-group.name }}
+        if: steps.changes.outputs.decision != 'skip'
         env:
           TEST_PATH: ${{ matrix.test-group.path }}
         run: |

litellm-proxy-extras/litellm_proxy_extras/migrations/20260710000000_add_dcr_bridge_to_mcp_server_table/migration.sql

@@ -0,0 +1,2 @@
+-- AlterTable
+ALTER TABLE "LiteLLM_MCPServerTable" ADD COLUMN     "dcr_bridge" BOOLEAN;

litellm-proxy-extras/litellm_proxy_extras/schema.prisma

@@ -339,6 +339,7 @@ model LiteLLM_MCPServerTable {
   available_on_public_internet Boolean @default(true)
   delegate_auth_to_upstream Boolean @default(false)
   oauth_passthrough Boolean @default(false)
+  dcr_bridge Boolean?
   is_byok               Boolean  @default(false)
   byok_description      String[] @default([])
   byok_api_key_help_url String?

litellm/constants.py

@@ -715,6 +715,7 @@ openai_compatible_endpoints: List = [
     "https://api.clarifai.com/v2/ext/openai/v1",
     "https://api.libertai.io/v1",
     "https://pinstripes.io/v1",
+    "https://api.meta.ai/v1",
 ]
 
 
@@ -781,6 +782,7 @@ openai_compatible_providers: List = [
     "ragflow",
     "pinstripes",  # Pinstripes - JSON-configured provider
     "darkbloom",
+    "meta",  # Meta Model API (Muse Spark) - JSON-configured provider
 ]
 openai_text_completion_compatible_providers: List = [  # providers that support `/v1/completions`
     "together_ai",

litellm/integrations/custom_guardrail.py

@@ -757,6 +757,12 @@ class CustomGuardrail(CustomLogger):
         # raw provider JSON so redaction is not duplicated upstream).
         clean_guardrail_response = redact_nested_match_and_regex_keys(clean_guardrail_response)
 
+        from litellm.litellm_core_utils.sensitive_data_masker import (
+            mask_credentials_in_payload,
+        )
+
+        clean_guardrail_response = mask_credentials_in_payload(clean_guardrail_response)
+
         slg = StandardLoggingGuardrailInformation(
             guardrail_name=self.guardrail_name,
             guardrail_provider=guardrail_provider,

litellm/litellm_core_utils/get_llm_provider_logic.py

@@ -346,6 +346,9 @@ def get_llm_provider(
                     elif endpoint == "https://pinstripes.io/v1":
                         custom_llm_provider = "pinstripes"
                         dynamic_api_key = get_secret_str("PINSTRIPES_API_KEY")
+                    elif endpoint == "https://api.meta.ai/v1":
+                        custom_llm_provider = "meta"
+                        dynamic_api_key = get_secret_str("META_API_KEY")
 
                     if api_base is not None and not isinstance(api_base, str):
                         raise Exception("api base needs to be a string. api_base={}".format(api_base))

litellm/litellm_core_utils/sensitive_data_masker.py

@@ -1,6 +1,8 @@
 from collections.abc import Mapping
 from typing import Any, Dict, List, Optional, Set
 
+from pydantic import BaseModel
+
 from litellm.constants import DEFAULT_MAX_RECURSE_DEPTH_SENSITIVE_DATA_MASKER
 
 
@@ -153,6 +155,39 @@ def mask_sensitive_structure(data: object) -> object:
     return _error_masker.mask(data)
 
 
+def mask_credentials_in_payload(data: object) -> object:
+    """Return a copy of ``data`` where string values under sensitive-named keys
+    are masked but every other value (``None``, ``int``, ``float``, ``bool``,
+    ``bytes``, ``datetime``, tuples, sets, typed objects) is preserved by
+    identity, and dicts/lists are rebuilt structurally.
+
+    Use this for logging payloads that carry response data through to
+    SpendLogs / OTel / Langfuse, where :meth:`SensitiveDataMasker.mask`'s
+    config-dump semantics (``None`` -> ``"None"``, tuples stringified,
+    objects flattened via ``__dict__``) would silently distort the record.
+
+    Sensitive-key detection is delegated to the shared
+    :class:`SensitiveDataMasker` so pattern updates stay in one place.
+    """
+    return _walk_payload(data, key_is_sensitive=False, depth=0)
+
+
+def _walk_payload(node: object, key_is_sensitive: bool, depth: int) -> object:
+    if depth >= DEFAULT_MAX_RECURSE_DEPTH_SENSITIVE_DATA_MASKER:
+        return node
+    if isinstance(node, Mapping):
+        return {k: _walk_payload(v, _default_masker.is_sensitive_key(k), depth + 1) for k, v in node.items()}
+    if isinstance(node, list):
+        return [_walk_payload(item, key_is_sensitive, depth + 1) for item in node]
+    if isinstance(node, tuple):
+        return tuple(_walk_payload(item, key_is_sensitive, depth + 1) for item in node)
+    if isinstance(node, BaseModel):
+        return _walk_payload(node.model_dump(), key_is_sensitive, depth)
+    if key_is_sensitive and isinstance(node, str) and node:
+        return _default_masker._mask_value(node)
+    return node
+
+
 def mask_sensitive_keys(data: Dict[str, Any], sensitive_fields: Set[str]) -> Dict[str, Any]:
     """Return a new dict with values masked for keys listed in ``sensitive_fields``.
 

litellm/llms/openai_like/dynamic_config.py

@@ -91,7 +91,7 @@ def create_config_class(provider: SimpleProviderConfig):
         def get_supported_openai_params(self, model: str) -> list:
             """Get supported OpenAI params, excluding tool-related params for models
             that don't support function calling."""
-            from litellm.utils import supports_function_calling
+            from litellm.utils import supports_function_calling, supports_reasoning
 
             supported_params = super().get_supported_openai_params(model=model)
 
@@ -113,6 +113,10 @@ def create_config_class(provider: SimpleProviderConfig):
                     f"function calling — removed tool-related params from supported params."
                 )
 
+            _supports_reasoning = supports_reasoning(model=model, custom_llm_provider=provider.slug)
+            if _supports_reasoning and "reasoning_effort" not in supported_params:
+                supported_params.append("reasoning_effort")
+
             return supported_params
 
         def map_openai_params(

litellm/llms/openai_like/messages/transformation.py

@@ -1,8 +1,11 @@
 from typing import Any, Optional
 
+import litellm
 from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
+from litellm.llms.openai_like.json_loader import SimpleProviderConfig
+from litellm.secret_managers.main import get_secret_str
 
 DEFAULT_ANTHROPIC_API_VERSION = "2023-06-01"
 
@@ -67,3 +70,65 @@ class OpenAILikeAnthropicMessagesConfig(AnthropicMessagesConfig):
         if base.endswith("/v1"):
             base = base[: -len("/v1")]
         return f"{base}/v1/messages"
+
+
+class JSONProviderAnthropicMessagesConfig(OpenAILikeAnthropicMessagesConfig):
+    """
+    Provider-level native Anthropic Messages passthrough for JSON-configured
+    OpenAI-compatible providers whose ``supported_endpoints`` in providers.json
+    includes ``"/v1/messages"``. Resolves the api key and api base from the
+    provider's configured env vars, then forwards the Anthropic payload
+    untranslated like ``OpenAILikeAnthropicMessagesConfig``.
+    """
+
+    def __init__(self, provider: SimpleProviderConfig):
+        super().__init__()
+        self._provider = provider
+
+    def should_strip_billing_metadata(self) -> bool:
+        return True
+
+    def _resolve_api_key(self, api_key: Optional[str]) -> Optional[str]:
+        return api_key or get_secret_str(self._provider.api_key_env) or litellm.api_key
+
+    def _resolve_api_base(self, api_base: Optional[str]) -> str:
+        env_api_base = get_secret_str(self._provider.api_base_env) if self._provider.api_base_env else None
+        return api_base or env_api_base or self._provider.base_url
+
+    def validate_anthropic_messages_environment(
+        self,
+        headers: dict[str, str],
+        model: str,
+        messages: list[Any],
+        optional_params: dict,
+        litellm_params: dict,
+        api_key: Optional[str] = None,
+        api_base: Optional[str] = None,
+    ) -> tuple[dict[str, str], Optional[str]]:
+        return super().validate_anthropic_messages_environment(
+            headers=headers,
+            model=model,
+            messages=messages,
+            optional_params=optional_params,
+            litellm_params=litellm_params,
+            api_key=self._resolve_api_key(api_key),
+            api_base=api_base,
+        )
+
+    def get_complete_url(
+        self,
+        api_base: Optional[str],
+        api_key: Optional[str],
+        model: str,
+        optional_params: dict,
+        litellm_params: dict,
+        stream: Optional[bool] = None,
+    ) -> str:
+        return super().get_complete_url(
+            api_base=self._resolve_api_base(api_base),
+            api_key=api_key,
+            model=model,
+            optional_params=optional_params,
+            litellm_params=litellm_params,
+            stream=stream,
+        )

litellm/llms/openai_like/providers.json

@@ -168,6 +168,13 @@
     },
     "supported_endpoints": ["/v1/chat/completions", "/v1/responses"]
   },
+  "meta": {
+    "base_url": "https://api.meta.ai/v1",
+    "api_key_env": "META_API_KEY",
+    "api_base_env": "META_API_BASE",
+    "base_class": "openai_gpt",
+    "supported_endpoints": ["/v1/chat/completions", "/v1/responses", "/v1/messages"]
+  },
   "pinstripes": {
     "base_url": "https://pinstripes.io/v1",
     "api_key_env": "PINSTRIPES_API_KEY",

litellm/model_prices_and_context_window_backup.json

@@ -6012,6 +6012,522 @@
         "supports_vision": true,
         "supports_web_search": true
     },
+    "azure/gpt-5.6": {
+        "cache_read_input_token_cost": 5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1e-06,
+        "cache_read_input_token_cost_priority": 1e-06,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 2e-06,
+        "input_cost_per_token": 5e-06,
+        "input_cost_per_token_above_272k_tokens": 1e-05,
+        "input_cost_per_token_priority": 1e-05,
+        "input_cost_per_token_above_272k_tokens_priority": 2e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.5e-05,
+        "output_cost_per_token_priority": 6e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 9e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/gpt-5.6-sol": {
+        "cache_read_input_token_cost": 5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1e-06,
+        "cache_read_input_token_cost_priority": 1e-06,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 2e-06,
+        "input_cost_per_token": 5e-06,
+        "input_cost_per_token_above_272k_tokens": 1e-05,
+        "input_cost_per_token_priority": 1e-05,
+        "input_cost_per_token_above_272k_tokens_priority": 2e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.5e-05,
+        "output_cost_per_token_priority": 6e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 9e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/gpt-5.6-terra": {
+        "cache_read_input_token_cost": 2.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 5e-07,
+        "cache_read_input_token_cost_priority": 5e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 1e-06,
+        "input_cost_per_token": 2.5e-06,
+        "input_cost_per_token_above_272k_tokens": 5e-06,
+        "input_cost_per_token_priority": 5e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 1e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.5e-05,
+        "output_cost_per_token_above_272k_tokens": 2.25e-05,
+        "output_cost_per_token_priority": 3e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 4.5e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/gpt-5.6-luna": {
+        "cache_read_input_token_cost": 1e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 2e-07,
+        "cache_read_input_token_cost_priority": 2e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 4e-07,
+        "input_cost_per_token": 1e-06,
+        "input_cost_per_token_above_272k_tokens": 2e-06,
+        "input_cost_per_token_priority": 2e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 4e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 6e-06,
+        "output_cost_per_token_above_272k_tokens": 9e-06,
+        "output_cost_per_token_priority": 1.2e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 1.8e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/us/gpt-5.6": {
+        "cache_read_input_token_cost": 5.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1.1e-06,
+        "cache_read_input_token_cost_priority": 1.375e-06,
+        "input_cost_per_token": 5.5e-06,
+        "input_cost_per_token_above_272k_tokens": 1.1e-05,
+        "input_cost_per_token_priority": 1.375e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.95e-05,
+        "output_cost_per_token_priority": 8.25e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/us/gpt-5.6-sol": {
+        "cache_read_input_token_cost": 5.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1.1e-06,
+        "cache_read_input_token_cost_priority": 1.375e-06,
+        "input_cost_per_token": 5.5e-06,
+        "input_cost_per_token_above_272k_tokens": 1.1e-05,
+        "input_cost_per_token_priority": 1.375e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.95e-05,
+        "output_cost_per_token_priority": 8.25e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/us/gpt-5.6-terra": {
+        "cache_read_input_token_cost": 2.75e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 5.5e-07,
+        "cache_read_input_token_cost_priority": 6.875e-07,
+        "input_cost_per_token": 2.75e-06,
+        "input_cost_per_token_above_272k_tokens": 5.5e-06,
+        "input_cost_per_token_priority": 6.875e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.65e-05,
+        "output_cost_per_token_above_272k_tokens": 2.475e-05,
+        "output_cost_per_token_priority": 4.125e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/us/gpt-5.6-luna": {
+        "cache_read_input_token_cost": 1.1e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 2.2e-07,
+        "cache_read_input_token_cost_priority": 2.75e-07,
+        "input_cost_per_token": 1.1e-06,
+        "input_cost_per_token_above_272k_tokens": 2.2e-06,
+        "input_cost_per_token_priority": 2.75e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 6.6e-06,
+        "output_cost_per_token_above_272k_tokens": 9.9e-06,
+        "output_cost_per_token_priority": 1.65e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/eu/gpt-5.6": {
+        "cache_read_input_token_cost": 5.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1.1e-06,
+        "cache_read_input_token_cost_priority": 1.375e-06,
+        "input_cost_per_token": 5.5e-06,
+        "input_cost_per_token_above_272k_tokens": 1.1e-05,
+        "input_cost_per_token_priority": 1.375e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.95e-05,
+        "output_cost_per_token_priority": 8.25e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/eu/gpt-5.6-sol": {
+        "cache_read_input_token_cost": 5.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1.1e-06,
+        "cache_read_input_token_cost_priority": 1.375e-06,
+        "input_cost_per_token": 5.5e-06,
+        "input_cost_per_token_above_272k_tokens": 1.1e-05,
+        "input_cost_per_token_priority": 1.375e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.95e-05,
+        "output_cost_per_token_priority": 8.25e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/eu/gpt-5.6-terra": {
+        "cache_read_input_token_cost": 2.75e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 5.5e-07,
+        "cache_read_input_token_cost_priority": 6.875e-07,
+        "input_cost_per_token": 2.75e-06,
+        "input_cost_per_token_above_272k_tokens": 5.5e-06,
+        "input_cost_per_token_priority": 6.875e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.65e-05,
+        "output_cost_per_token_above_272k_tokens": 2.475e-05,
+        "output_cost_per_token_priority": 4.125e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/eu/gpt-5.6-luna": {
+        "cache_read_input_token_cost": 1.1e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 2.2e-07,
+        "cache_read_input_token_cost_priority": 2.75e-07,
+        "input_cost_per_token": 1.1e-06,
+        "input_cost_per_token_above_272k_tokens": 2.2e-06,
+        "input_cost_per_token_priority": 2.75e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 6.6e-06,
+        "output_cost_per_token_above_272k_tokens": 9.9e-06,
+        "output_cost_per_token_priority": 1.65e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
     "azure/gpt-5.5": {
         "cache_read_input_token_cost": 5e-07,
         "cache_read_input_token_cost_above_272k_tokens": 1e-06,
@@ -25501,6 +26017,42 @@
         "supports_function_calling": true,
         "supports_tool_choice": false
     },
+    "meta/muse-spark-1.1": {
+        "cache_read_input_token_cost": 1.5e-07,
+        "input_cost_per_token": 1.25e-06,
+        "litellm_provider": "meta",
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 131072,
+        "max_tokens": 131072,
+        "mode": "chat",
+        "output_cost_per_token": 4.25e-06,
+        "source": "https://dev.meta.ai/docs/getting-started/pricing-rate-limits",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/responses",
+            "/v1/messages"
+        ],
+        "supported_modalities": [
+            "text",
+            "image",
+            "video"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_minimal_reasoning_effort": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_xhigh_reasoning_effort": true
+    },
     "meta_llama/Llama-3.3-70B-Instruct": {
         "litellm_provider": "meta_llama",
         "max_input_tokens": 128000,

litellm/models/mcp_server.py

@@ -96,6 +96,7 @@ class LiteLLM_MCPServerTable(LiteLLMPydanticObjectBase):
     available_on_public_internet: bool = True
     delegate_auth_to_upstream: bool = False
     oauth_passthrough: bool = False
+    dcr_bridge: Optional[bool] = None
     is_byok: bool = False
     byok_description: List[str] = Field(default_factory=list)
     byok_api_key_help_url: Optional[str] = None

litellm/proxy/_experimental/mcp_server/db.py

@@ -3,7 +3,7 @@ import binascii
 import hashlib
 import json
 from datetime import datetime, timedelta, timezone
-from typing import TYPE_CHECKING, Any, Dict, Iterable, List, Optional, Set, Union, cast
+from typing import TYPE_CHECKING, Any, Awaitable, Callable, Dict, Iterable, List, Optional, Set, Union, cast
 
 from litellm._logging import verbose_proxy_logger
 from litellm._uuid import uuid
@@ -52,6 +52,7 @@ _AUTH_FLOW_SCOPED_FIELDS: frozenset = frozenset(
         "token_url",
         "registration_url",
         "oauth2_flow",
+        "dcr_bridge",
         "token_exchange_endpoint",
         "audience",
         "subject_token_type",
@@ -558,7 +559,11 @@ async def delete_mcp_server_from_virtualkey():
     pass
 
 
-async def delete_mcp_server(prisma_client: PrismaClient, server_id: str) -> Optional[LiteLLM_MCPServerTable]:
+async def delete_mcp_server(
+    prisma_client: PrismaClient,
+    server_id: str,
+    invalidate_token_cache: Optional[Callable[[str, str], Awaitable[None]]] = None,
+) -> Optional[LiteLLM_MCPServerTable]:
     """
     Delete the mcp server from the db by server_id
 
@@ -569,6 +574,12 @@ async def delete_mcp_server(prisma_client: PrismaClient, server_id: str) -> Opti
     caller-visible error. Each table is cleaned independently so a failure on one
     still attempts the other.
 
+    Each enumerated credential row's user also gets their cached per-user token
+    invalidated (legacy cache + v2 store, via invalidate_token_cache, defaulting
+    to the manager's shared invalidation): the caches are keyed by
+    (user_id, server_id), so without this a re-created server reusing the same
+    server_id would serve tokens minted for the deleted server until TTL.
+
     Returns the deleted mcp server record if it exists, otherwise None
     """
     deleted_server = await MCPServerRepository(prisma_client).table.delete(
@@ -577,6 +588,18 @@ async def delete_mcp_server(prisma_client: PrismaClient, server_id: str) -> Opti
         },
     )
     if deleted_server is not None:
+        credential_user_ids: List[str] = []
+        try:
+            credential_rows = await prisma_client.db.litellm_mcpusercredentials.find_many(
+                where={"server_id": server_id}
+            )
+            credential_user_ids = [row.user_id for row in credential_rows]
+        except Exception as e:  # noqa: BLE001 - enumeration is best-effort; cached tokens expire by TTL
+            verbose_proxy_logger.warning(
+                "MCP server %s deleted but per-user credential enumeration failed; cached tokens expire by TTL: %s",
+                server_id,
+                e,
+            )
         for model, label in (
             (prisma_client.db.litellm_mcpusercredentials, "credential"),
             (prisma_client.db.litellm_mcpuserenvvars, "env var"),
@@ -591,6 +614,15 @@ async def delete_mcp_server(prisma_client: PrismaClient, server_id: str) -> Opti
                     label,
                     e,
                 )
+        if credential_user_ids:
+            if invalidate_token_cache is None:
+                from litellm.proxy._experimental.mcp_server.mcp_server_manager import (
+                    global_mcp_server_manager,
+                )
+
+                invalidate_token_cache = global_mcp_server_manager.invalidate_user_oauth_token_cache
+            for user_id in credential_user_ids:
+                await invalidate_token_cache(user_id, server_id)
     return deleted_server
 
 
@@ -1070,6 +1102,103 @@ async def list_user_oauth_credentials(
     return results
 
 
+def _decrypted_credential_field(creds: Dict[str, object], field: str) -> object:
+    """Return one credential field decrypted with the global salt key; non-string and legacy
+    plaintext values come back unchanged (decrypt_value_helper returns the original on failure)."""
+    value = creds.get(field)
+    if not isinstance(value, str):
+        return value
+    return decrypt_value_helper(
+        value=value,
+        key=field,
+        exception_type="debug",
+        return_original_value=True,
+    )
+
+
+def mcp_oauth_token_identity(server: object) -> tuple[object, ...]:
+    """The upstream-OAuth-token-determining fields of an MCP server: the resource/audience (url, or
+    spec_path for OpenAPI servers), the OAuth mode/grant (auth_type, oauth2_flow), the
+    authorization-server endpoints, and the OAuth client + scopes. Mirrors the dashboard's
+    getOAuthAuthorizationIdentity. When any of these change on a server update, previously stored
+    per-user tokens were minted for the old identity and are stale. Excludes transport and
+    delegate_auth_to_upstream, which do not affect what token is minted (RFC 8707/8693).
+
+    client_id/client_secret are compared decrypted: stored values are NaCl-encrypted with a fresh
+    nonce on every write, so comparing ciphertext would flag every routine save as an identity
+    change and purge tokens that are still valid."""
+    creds = getattr(server, "credentials", None)
+    if isinstance(creds, str):
+        try:
+            parsed: object = json.loads(creds)
+        except ValueError:
+            parsed = None
+    else:
+        parsed = creds
+    creds_dict: Dict[str, object] = parsed if isinstance(parsed, dict) else {}
+    return (
+        getattr(server, "url", None),
+        getattr(server, "spec_path", None),
+        getattr(server, "auth_type", None),
+        getattr(server, "oauth2_flow", None),
+        getattr(server, "authorization_url", None),
+        getattr(server, "token_url", None),
+        getattr(server, "registration_url", None),
+        _decrypted_credential_field(creds_dict, "client_id"),
+        _decrypted_credential_field(creds_dict, "client_secret"),
+        creds_dict.get("scopes"),
+    )
+
+
+async def purge_user_oauth_credentials_for_server(
+    prisma_client: PrismaClient,
+    server_id: str,
+    invalidate_token_cache: Optional[Callable[[str, str], Awaitable[None]]] = None,
+) -> int:
+    """Delete every stored per-user OAuth token for a server and invalidate each user's cached
+    token everywhere it can be served from (the legacy per-user token cache and the v2 per-user OAuth
+    token store), so no user keeps a token minted for a superseded configuration. Called when a server
+    update changes a mint-relevant field (see mcp_oauth_token_identity). Returns the number of rows
+    removed.
+
+    LiteLLM_MCPUserCredentials also stores BYOK API keys in the same column; only rows whose payload
+    decodes as an OAuth2 credential (see _decode_oauth_payload) are deleted, because a config change
+    only invalidates minted tokens, never a user's own stored key. Rows are therefore deleted per
+    (user_id, server_id) pair rather than by a blanket server_id filter. An OAuth row inserted while
+    the purge runs for a user not yet enumerated survives; a re-auth completing in the window for an
+    already-enumerated user is deleted along with the stale row (the pair delete cannot tell them
+    apart), which costs that user one extra re-auth and nothing else.
+
+    invalidate_token_cache is injectable for tests; it defaults to the manager's shared
+    invalidate_user_oauth_token_cache, the single invalidation point for per-user tokens."""
+    repo = MCPUserCredentialsRepository(prisma_client)
+    rows = await repo.table.find_many(where={"server_id": server_id})
+    oauth_rows = [row for row in rows if _decode_oauth_payload(row.credential_b64) is not None]
+    if not oauth_rows:
+        return 0
+    deleted_count = await repo.table.delete_many(
+        where={"server_id": server_id, "user_id": {"in": [row.user_id for row in oauth_rows]}}
+    )
+    if invalidate_token_cache is None:
+        from litellm.proxy._experimental.mcp_server.mcp_server_manager import (
+            global_mcp_server_manager,
+        )
+
+        invalidate_token_cache = global_mcp_server_manager.invalidate_user_oauth_token_cache
+
+    for row in oauth_rows:
+        await invalidate_token_cache(row.user_id, server_id)
+    if deleted_count != len(oauth_rows):
+        verbose_proxy_logger.warning(
+            "MCP server %s: purge removed %d OAuth credential row(s) but %d were enumerated; "
+            "row(s) were deleted concurrently during the purge",
+            server_id,
+            deleted_count,
+            len(oauth_rows),
+        )
+    return deleted_count
+
+
 async def refresh_user_oauth_token(
     prisma_client: PrismaClient,
     user_id: str,

litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py

@@ -471,7 +471,8 @@ def _raise_if_not_oauth2(mcp_server: MCPServer) -> None:
     through: the caller owns the upstream token, and this relayed flow is how a browser obtains
     one against the upstream IdP (the admin UI's browser-only Authorize uses it). The minted
     token is upstream-audienced and held by the caller; the gateway persists nothing for these
-    modes (DCR persistence is opt-in and never enabled on this path).
+    modes (``_persist_dcr_client_registration`` skips them unconditionally, so even the admin
+    Authorize path with ``persist_credentials`` enabled writes nothing to the server row).
     """
     from litellm.proxy._experimental.mcp_server.mcp_server_manager import (  # noqa: PLC0415  # circular import with mcp_server_manager at module load
         _UPSTREAM_OAUTH_DISCOVERY_AUTH_TYPES,
@@ -807,7 +808,7 @@ async def _reuse_persisted_dcr_client_if_available(mcp_server: MCPServer) -> boo
     return bool(mcp_server.client_id)
 
 
-DcrRegistrationPersistenceResult = Literal["persisted", "reused", "failed"]
+DcrRegistrationPersistenceResult = Literal["persisted", "reused", "skipped", "failed"]
 
 
 async def _persist_dcr_client_registration(
@@ -821,7 +822,16 @@ async def _persist_dcr_client_registration(
     full re-authorization instead of a silent refresh. Mirrors the ``encrypt_credentials``
     write that ``client_credentials`` and token exchange already use. Failures are logged,
     never raised: registration still returns to the caller even when persistence fails.
+
+    The client-forwarded token modes (``true_passthrough`` / ``oauth_delegate``) are skipped
+    unconditionally: the caller holds the upstream token and the gateway must hold no OAuth
+    client identity for these servers. Persisting here would stamp ``oauth2_flow`` and a
+    ``client_id`` onto a server whose mode promises the gateway stores nothing, making a
+    fresh pass-through server read as gateway-authorized.
     """
+    if mcp_server.is_true_passthrough or mcp_server.is_oauth_delegate:
+        return "skipped"
+
     try:
         registration = _DcrClientRegistration.model_validate(registration_response)
     except ValidationError as exc:

litellm/proxy/_experimental/mcp_server/mcp_server_manager.py

@@ -57,7 +57,11 @@ from litellm.proxy._experimental.mcp_server.elicitation_handler import (
 from litellm.proxy._experimental.mcp_server.sampling_handler import (
     MCP_SAMPLING_AVAILABLE,
 )
-from litellm.proxy._experimental.mcp_server.oauth2_token_cache import resolve_mcp_auth
+from litellm.proxy._experimental.mcp_server.oauth2_token_cache import (
+    MCPPerUserTokenCache,
+    mcp_per_user_token_cache,
+    resolve_mcp_auth,
+)
 from litellm.proxy._experimental.mcp_server.outbound_credentials import (
     Error,
     Ok,
@@ -799,10 +803,12 @@ class MCPServerManager:
         self,
         cred_provider: Optional[UpstreamCredentialProvider] = None,
         per_user_oauth_token_store: Optional[InvalidatableOAuthTokenStore] = None,
+        per_user_token_cache: Optional[MCPPerUserTokenCache] = None,
     ):
         self._per_user_oauth_token_store = per_user_oauth_token_store or LazyPerUserOAuthTokenStore(
             self.get_mcp_server_by_id
         )
+        self._per_user_token_cache = per_user_token_cache or mcp_per_user_token_cache
         self._cred_provider = cred_provider or UpstreamCredentialProvider(
             oauth_token_store=self._per_user_oauth_token_store,
             token_exchanger=build_token_exchanger(),
@@ -1038,6 +1044,24 @@ class MCPServerManager:
                     "browser sign-in, including delegate_auth_to_upstream)."
                 )
 
+            config_dcr_bridge = server_config.get("dcr_bridge", None)
+            if config_dcr_bridge is not None and not isinstance(config_dcr_bridge, bool):
+                raise ValueError(
+                    f"Invalid config for MCP server '{server_name or server_id}': dcr_bridge "
+                    f"must be a boolean (got {config_dcr_bridge!r})."
+                )
+            if config_dcr_bridge and auth_type not in (
+                MCPAuth.true_passthrough,
+                MCPAuth.oauth_delegate,
+            ):
+                raise ValueError(
+                    f"Invalid config for MCP server '{server_name or server_id}': dcr_bridge is only "
+                    f"supported for auth_type true_passthrough or oauth_delegate (got {auth_type!r}). "
+                    "The DCR bridge serves gateway-hosted OAuth discovery for the client-forwarded "
+                    "token modes; interactive oauth2 servers already run the gateway "
+                    "authorization-code flow."
+                )
+
             new_server = MCPServer(
                 server_id=server_id,
                 name=name_for_prefix,
@@ -1073,6 +1097,7 @@ class MCPServerManager:
                 available_on_public_internet=bool(server_config.get("available_on_public_internet", True)),
                 delegate_auth_to_upstream=bool(server_config.get("delegate_auth_to_upstream", False)),
                 oauth_passthrough=bool(server_config.get("oauth_passthrough", False)),
+                dcr_bridge=config_dcr_bridge,
                 # AWS SigV4 fields
                 aws_access_key_id=server_config.get("aws_access_key_id", None),
                 aws_secret_access_key=server_config.get("aws_secret_access_key", None),
@@ -1448,6 +1473,7 @@ class MCPServerManager:
             available_on_public_internet=bool(getattr(mcp_server, "available_on_public_internet", True)),
             delegate_auth_to_upstream=bool(getattr(mcp_server, "delegate_auth_to_upstream", False)),
             oauth_passthrough=bool(getattr(mcp_server, "oauth_passthrough", False)),
+            dcr_bridge=getattr(mcp_server, "dcr_bridge", None),
             created_at=getattr(mcp_server, "created_at", None),
             updated_at=getattr(mcp_server, "updated_at", None),
             tool_name_to_display_name=_deserialize_json_dict(getattr(mcp_server, "tool_name_to_display_name", None)),
@@ -4053,10 +4079,13 @@ class MCPServerManager:
         return await self._cred_provider.has_user_token(to_subject(user_api_key_auth, None), spec)
 
     async def invalidate_user_oauth_token_cache(self, user_id: str, server_id: str) -> None:
-        """Drop the v2 chain's cached token for ``(user_id, server_id)`` after the credential row
-        changes (re-auth, revoke), so the next resolve reads the new row instead of serving the
-        replaced token until its cache TTL. Best-effort: a cache-drop failure is logged, never
-        raised, because the DB write already succeeded and the TTL remains the backstop.
+        """Drop every cached token for ``(user_id, server_id)`` after the credential row changes
+        (re-auth, revoke, config-change purge): the v2 chain's cache and the legacy per-user token
+        cache, so the next resolve reads the new row instead of serving the replaced token until its
+        cache TTL, whichever path resolves it. This is the single invalidation point for per-user
+        OAuth tokens; callers must not evict individual caches directly. Best-effort: a cache-drop
+        failure is logged, never raised, because the DB write already succeeded and the TTL remains
+        the backstop.
         """
         try:
             await self._per_user_oauth_token_store.invalidate(user_id, server_id)
@@ -4064,6 +4093,12 @@ class MCPServerManager:
             verbose_logger.warning(
                 "Failed to invalidate cached MCP OAuth token for user=%s server=%s: %s", user_id, server_id, exc
             )
+        try:
+            await self._per_user_token_cache.delete(user_id, server_id)
+        except Exception as exc:  # noqa: BLE001 - cache drop is best-effort; TTL is the backstop
+            verbose_logger.warning(
+                "Failed to drop legacy cached MCP OAuth token for user=%s server=%s: %s", user_id, server_id, exc
+            )
 
     async def _resolve_oauth2_headers_for_tool_call(
         self,
@@ -4785,6 +4820,7 @@ class MCPServerManager:
             token_url=server.token_url,
             registration_url=server.registration_url,
             oauth2_flow=server.oauth2_flow,
+            dcr_bridge=server.dcr_bridge,
             token_exchange_endpoint=server.token_exchange_endpoint,
             audience=server.audience,
             subject_token_type=server.subject_token_type,
@@ -4901,6 +4937,7 @@ class MCPServerManager:
             available_on_public_internet=server.available_on_public_internet,
             delegate_auth_to_upstream=server.delegate_auth_to_upstream,
             oauth_passthrough=getattr(server, "oauth_passthrough", False),
+            dcr_bridge=server.dcr_bridge,
             is_byok=server.is_byok,
             byok_description=server.byok_description,
             byok_api_key_help_url=server.byok_api_key_help_url,

litellm/proxy/_types.py

@@ -26,6 +26,7 @@ from litellm.types.llms.openai import (
     ResponsesAPIResponse,
 )
 from litellm.types.mcp import (
+    MCPAuth,
     MCPAuthType,
     MCPCredentials,
     MCPTransport,
@@ -1229,6 +1230,14 @@ from litellm.models.mcp_server import (  # noqa: E402
 
 
 # MCP Proxy Request Types
+def _dcr_bridge_auth_type_error(auth_type: object) -> ValueError:
+    return ValueError(
+        f"dcr_bridge is only supported for auth_type true_passthrough or oauth_delegate (got {auth_type!r}). "
+        "The DCR bridge serves gateway-hosted OAuth discovery for the client-forwarded token modes; "
+        "interactive oauth2 servers already run the gateway authorization-code flow."
+    )
+
+
 class NewMCPServerRequest(LiteLLMPydanticObjectBase):
     server_id: Optional[str] = None
     server_name: Optional[str] = None
@@ -1268,6 +1277,7 @@ class NewMCPServerRequest(LiteLLMPydanticObjectBase):
     available_on_public_internet: bool = True
     delegate_auth_to_upstream: bool = False
     oauth_passthrough: bool = False
+    dcr_bridge: Optional[bool] = None
     is_byok: bool = False
     byok_description: List[str] = Field(default_factory=list)
     byok_api_key_help_url: Optional[str] = None
@@ -1322,6 +1332,16 @@ class NewMCPServerRequest(LiteLLMPydanticObjectBase):
         """
         return values
 
+    @model_validator(mode="before")
+    @classmethod
+    def validate_dcr_bridge_auth_type(cls, values):
+        if not isinstance(values, dict) or not values.get("dcr_bridge"):
+            return values
+        auth_type = values.get("auth_type")
+        if auth_type in (MCPAuth.true_passthrough, MCPAuth.oauth_delegate):
+            return values
+        raise _dcr_bridge_auth_type_error(auth_type)
+
 
 class UpdateMCPServerRequest(LiteLLMPydanticObjectBase):
     server_id: str
@@ -1362,6 +1382,7 @@ class UpdateMCPServerRequest(LiteLLMPydanticObjectBase):
     available_on_public_internet: bool = True
     delegate_auth_to_upstream: bool = False
     oauth_passthrough: bool = False
+    dcr_bridge: Optional[bool] = None
     is_byok: bool = False
     byok_description: List[str] = Field(default_factory=list)
     byok_api_key_help_url: Optional[str] = None
@@ -1391,6 +1412,21 @@ class UpdateMCPServerRequest(LiteLLMPydanticObjectBase):
                     raise ValueError("url or spec_path is required for HTTP/SSE transport")
         return values
 
+    @model_validator(mode="before")
+    @classmethod
+    def validate_dcr_bridge_auth_type(cls, values):
+        """Partial updates omit auth_type; that case is validated against the stored row by the
+        update endpoint, which can read the database. This validator covers payloads that carry
+        both fields."""
+        if not isinstance(values, dict) or not values.get("dcr_bridge"):
+            return values
+        if "auth_type" not in values:
+            return values
+        auth_type = values.get("auth_type")
+        if auth_type in (MCPAuth.true_passthrough, MCPAuth.oauth_delegate):
+            return values
+        raise _dcr_bridge_auth_type_error(auth_type)
+
 
 from litellm.models.mcp_server import (  # noqa: E402
     LiteLLM_MCPServerTable as LiteLLM_MCPServerTable,

litellm/proxy/auth/auth_checks.py

@@ -98,7 +98,10 @@ from litellm.repositories.user_repository import UserRepository
 from litellm.router import Router
 from litellm.utils import get_utc_datetime
 
-from .auth_checks_organization import organization_role_based_access_check
+from .auth_checks_organization import (
+    add_team_org_context_to_request_body,
+    organization_role_based_access_check,
+)
 from .auth_utils import get_model_from_request
 
 if TYPE_CHECKING:
@@ -707,10 +710,28 @@ async def common_checks(
     # 10 [OPTIONAL] Organization RBAC checks
     organization_role_based_access_check(user_object=user_object, route=route, request_body=request_body)
 
+    async def _fetch_team_org_id(team_id: str) -> Optional[str]:
+        try:
+            team = await get_team_object(
+                team_id=team_id,
+                prisma_client=prisma_client,
+                user_api_key_cache=user_api_key_cache,
+                proxy_logging_obj=proxy_logging_obj,
+            )
+        except HTTPException:
+            return None
+        return team.organization_id
+
+    request_body_for_route_check = await add_team_org_context_to_request_body(
+        route=route,
+        request_body=request_body,
+        fetch_team_org_id=_fetch_team_org_id,
+    )
+
     _is_route_allowed = _is_api_route_allowed(
         route=route,
         request=request,
-        request_data=request_body,
+        request_data=request_body_for_route_check,
         valid_token=valid_token,
         user_obj=user_object,
     )

litellm/proxy/auth/auth_checks_organization.py

@@ -2,7 +2,7 @@
 Auth Checks for Organizations
 """
 
-from typing import Dict, List, Optional, Tuple
+from typing import Awaitable, Callable, Dict, List, Optional, Tuple
 
 from fastapi import status
 
@@ -170,3 +170,33 @@ def _user_is_org_admin(
 
     # User must be admin of ALL requested orgs, not just any one
     return all(org_id in admin_org_ids for org_id in candidate_org_ids)
+
+
+TEAM_ORG_CONTEXT_ROUTES = frozenset({"/team/update"})
+
+
+async def add_team_org_context_to_request_body(
+    route: str,
+    request_body: dict,
+    fetch_team_org_id: Callable[[str], Awaitable[Optional[str]]],
+) -> dict:
+    """
+    Return a copy of request_body with organization_id resolved from the target
+    team when the route identifies the team by team_id and the caller did not
+    pass organization_id. This lets an org admin of the team's own org reach the
+    org-scoped branch of the route gate (which keys off organization_id) without
+    the client having to send it. Returns request_body unchanged when it does
+    not apply, so callers that already pass organization_id and non-team routes
+    are untouched.
+    """
+    if route not in TEAM_ORG_CONTEXT_ROUTES:
+        return request_body
+    if request_body.get("organization_id"):
+        return request_body
+    team_id = request_body.get("team_id")
+    if not isinstance(team_id, str) or not team_id:
+        return request_body
+    org_id = await fetch_team_org_id(team_id)
+    if not org_id:
+        return request_body
+    return {**request_body, "organization_id": org_id}

litellm/proxy/management_endpoints/mcp_management_endpoints.py

@@ -125,7 +125,9 @@ if MCP_AVAILABLE:
         get_user_env_vars_bulk,
         get_user_oauth_credential,
         list_user_oauth_credentials,
+        mcp_oauth_token_identity,
         merge_user_env_vars,
+        purge_user_oauth_credentials_for_server,
         reject_mcp_server,
         store_user_credential,
         store_user_oauth_credential,
@@ -2318,6 +2320,41 @@ if MCP_AVAILABLE:
                 },
             )
 
+        # Snapshot the pre-update identity so we can detect a mint-relevant change below. The read is
+        # advisory (it only feeds the stale-token purge decision), so a failure skips the purge with a
+        # warning instead of failing the edit, whose primary job is the update itself.
+        try:
+            old_server_record = await get_mcp_server(prisma_client, payload.server_id)
+            old_server_record_read_failed = False
+        except Exception as exc:  # noqa: BLE001 - advisory read; invalidation is best-effort end-to-end
+            verbose_logger.warning(
+                "MCP server %s: could not snapshot the pre-update record; skipping the stale-token check: %s",
+                payload.server_id,
+                exc,
+            )
+            old_server_record = None
+            old_server_record_read_failed = True
+
+        if (
+            payload.dcr_bridge
+            and payload.auth_type is None
+            and (old_server_record is not None or old_server_record_read_failed)
+        ):
+            stored_auth_type = old_server_record.auth_type if old_server_record else None
+            stored_auth_type_name = getattr(stored_auth_type, "value", stored_auth_type)
+            if stored_auth_type not in (MCPAuth.true_passthrough, MCPAuth.oauth_delegate):
+                raise HTTPException(
+                    status_code=status.HTTP_400_BAD_REQUEST,
+                    detail={
+                        "error": (
+                            "dcr_bridge is only supported for auth_type true_passthrough or "
+                            f"oauth_delegate (stored auth_type: {stored_auth_type_name!r}). Include "
+                            "the server's auth_type in the update payload or configure one of the "
+                            "client-forwarded token modes first."
+                        )
+                    },
+                )
+
         # try to update the mcp server
         mcp_server_record_updated = await update_mcp_server(
             prisma_client,
@@ -2336,6 +2373,30 @@ if MCP_AVAILABLE:
         # Ensure registry is up to date by reloading from database
         await global_mcp_server_manager.reload_servers_from_database()
 
+        # If a field that determines which upstream OAuth token gets minted changed (url/audience, OAuth
+        # mode/grant, authorization-server endpoints, or the OAuth client + scopes), every stored per-user
+        # token was minted for the old configuration and is stale. Purge them (DB + cache) so the next
+        # tool call re-authorizes instead of forwarding a token for a resource/AS/client that no longer
+        # matches. Best-effort: a purge failure must not fail the update, whose primary job already
+        # succeeded.
+        if old_server_record is not None and mcp_oauth_token_identity(old_server_record) != mcp_oauth_token_identity(
+            mcp_server_record_updated
+        ):
+            try:
+                purged = await purge_user_oauth_credentials_for_server(prisma_client, payload.server_id)
+                if purged:
+                    verbose_logger.info(
+                        "MCP server %s: purged %d stale per-user OAuth token(s) after a mint-relevant config change",
+                        payload.server_id,
+                        purged,
+                    )
+            except Exception as exc:  # noqa: BLE001 - purge is best-effort; the server update already succeeded
+                verbose_logger.warning(
+                    "MCP server %s: failed to purge stale per-user OAuth tokens after config change: %s",
+                    payload.server_id,
+                    exc,
+                )
+
         # TODO: Enterprise: Finish audit log trail
         if litellm.store_audit_logs:
             pass

litellm/proxy/schema.prisma

@@ -339,6 +339,7 @@ model LiteLLM_MCPServerTable {
   available_on_public_internet Boolean @default(true)
   delegate_auth_to_upstream Boolean @default(false)
   oauth_passthrough Boolean @default(false)
+  dcr_bridge Boolean?
   is_byok               Boolean  @default(false)
   byok_description      String[] @default([])
   byok_api_key_help_url String?

litellm/proxy/spend_tracking/spend_management_endpoints.py

@@ -3392,7 +3392,7 @@ async def _build_ui_spend_logs_response(
             )
             count_map = {r["session_id"]: r["_count"]["session_id"] for r in counts if r.get("session_id")}
 
-    mcp_spend_map: dict[str, dict[str, Union[int, float]]] = {}
+    session_spend_map: dict[str, dict[str, Union[int, float]]] = {}
     if enrich_session_counts and session_ids:
         from prisma.errors import PrismaError
 
@@ -3410,19 +3410,24 @@ async def _build_ui_spend_logs_response(
             rows = await prisma_client.db.query_raw(
                 """
                 SELECT session_id,
-                       COUNT(*)::int AS mcp_tool_call_count,
-                       COALESCE(SUM(spend), 0)::double precision AS mcp_tool_call_spend
+                       COALESCE(SUM(spend), 0)::double precision AS session_total_spend,
+                       COUNT(*) FILTER (
+                           WHERE call_type IN ('call_mcp_tool', 'list_mcp_tools')
+                       )::int AS mcp_tool_call_count,
+                       COALESCE(SUM(spend) FILTER (
+                           WHERE call_type IN ('call_mcp_tool', 'list_mcp_tools')
+                       ), 0)::double precision AS mcp_tool_call_spend
                 FROM "LiteLLM_SpendLogs"
                 WHERE session_id = ANY($1::text[])
                   AND api_key = ANY($2::text[])
-                  AND call_type IN ('call_mcp_tool', 'list_mcp_tools')
                 GROUP BY session_id
                 """,
                 session_ids,
                 authorized_api_keys,
             )
-            mcp_spend_map = {
+            session_spend_map = {
                 row["session_id"]: {
+                    "session_total_spend": float(row.get("session_total_spend") or 0.0),
                     "mcp_tool_call_count": int(row.get("mcp_tool_call_count") or 0),
                     "mcp_tool_call_spend": float(row.get("mcp_tool_call_spend") or 0.0),
                 }
@@ -3431,7 +3436,7 @@ async def _build_ui_spend_logs_response(
             }
         except PrismaError:
             verbose_proxy_logger.debug(
-                "Failed to enrich MCP session spend aggregates for spend logs UI",
+                "Failed to enrich session spend aggregates for spend logs UI",
                 exc_info=True,
             )
 
@@ -3441,10 +3446,12 @@ async def _build_ui_spend_logs_response(
             row_dict = dict(row) if isinstance(row, dict) else row.model_dump()
             sid = row_dict.get("session_id")
             row_dict["session_total_count"] = count_map.get(sid, 1) if sid else 1
-            mcp_stats = mcp_spend_map.get(sid) if sid else None
-            if mcp_stats:
-                row_dict["mcp_tool_call_count"] = mcp_stats["mcp_tool_call_count"]
-                row_dict["mcp_tool_call_spend"] = mcp_stats["mcp_tool_call_spend"]
+            session_stats = session_spend_map.get(sid) if sid else None
+            if session_stats:
+                row_dict["session_total_spend"] = session_stats["session_total_spend"]
+                if session_stats["mcp_tool_call_count"]:
+                    row_dict["mcp_tool_call_count"] = session_stats["mcp_tool_call_count"]
+                    row_dict["mcp_tool_call_spend"] = session_stats["mcp_tool_call_spend"]
             enriched.append(row_dict)
         response_data: list = enriched
     else:

litellm/proxy/spend_tracking/spend_tracking_utils.py

@@ -136,7 +136,7 @@ def _get_spend_logs_metadata(
     clean_metadata["vector_store_request_metadata"] = _get_vector_store_request_for_spend_logs_payload(
         vector_store_request_metadata
     )
-    clean_metadata["guardrail_information"] = guardrail_information
+    clean_metadata["guardrail_information"] = _sanitize_guardrail_information_for_spend_logs(guardrail_information)
     clean_metadata["usage_object"] = usage_object
     clean_metadata["model_map_information"] = model_map_information
     clean_metadata["cold_storage_object_key"] = cold_storage_object_key
@@ -868,6 +868,51 @@ def _redact_prompt_leaks_in_error_string(text: str) -> str:
     return "".join(out)
 
 
+def _sanitize_guardrail_information_for_spend_logs(
+    guardrail_information: Optional[List[StandardLoggingGuardrailInformation]],
+) -> Optional[List[StandardLoggingGuardrailInformation]]:
+    """
+    When ``store_prompts_in_spend_logs`` is False, redact prompt-carrying fields
+    (``guardrail_request``, ``guardrail_response``, ``match_details``,
+    ``classification``) before they land in ``LiteLLM_SpendLogs.metadata``.
+
+    Guardrail hooks may echo the LLM request payload back into
+    ``guardrail_response``, and two first-party hooks
+    (``block_code_execution``, ``litellm_content_filter``) inline user-prompt
+    substrings into ``match_details`` / ``classification`` too, so the flag
+    must cover all four fields. Every other typed field on the entry (name,
+    provider, mode, status, timings, action, violation_categories, risk_score,
+    masked_entity_count, ...) is preserved so guardrail dashboards keep
+    working.
+
+    ``guardrail_information`` is typed ``Optional[List[...]]`` but at least
+    one writer (``xecguard``) assigns a bare dict, so normalize to a list
+    here to match OTEL's defensive read pattern; otherwise iteration would
+    yield the dict's keys and crash the whole spend-log write.
+    """
+    if guardrail_information is None or _should_store_prompts_and_responses_in_spend_logs():
+        return guardrail_information
+    entries = [guardrail_information] if isinstance(guardrail_information, dict) else guardrail_information
+    return [_redact_prompt_fields_in_guardrail_entry(entry) for entry in entries if isinstance(entry, dict)]
+
+
+_PROMPT_CARRYING_GUARDRAIL_FIELDS = (
+    "guardrail_request",
+    "guardrail_response",
+    "match_details",
+    "classification",
+)
+
+
+def _redact_prompt_fields_in_guardrail_entry(
+    entry: StandardLoggingGuardrailInformation,
+) -> StandardLoggingGuardrailInformation:
+    return {
+        **entry,
+        **{key: REDACTED_BY_LITELM_STRING for key in _PROMPT_CARRYING_GUARDRAIL_FIELDS if key in entry},
+    }
+
+
 def _sanitize_error_information_for_spend_logs(
     error_information: Optional[StandardLoggingPayloadErrorInformation],
 ) -> Optional[StandardLoggingPayloadErrorInformation]:

litellm/types/mcp_server/mcp_server_manager.py

@@ -103,6 +103,7 @@ class MCPServer(BaseModel):
     # ``Authorization`` for non-OAuth reasons (e.g. static bearer tokens). Must
     # be set explicitly to avoid regressing servers that did not opt in.
     oauth_passthrough: bool = False
+    dcr_bridge: Optional[bool] = None
     is_byok: bool = False
     byok_description: List[str] = []
     byok_api_key_help_url: Optional[str] = None
@@ -164,6 +165,15 @@ class MCPServer(BaseModel):
         JWT) but forwards the caller's separate upstream ``Authorization`` unchanged, minting nothing."""
         return self.auth_type == MCPAuth.oauth_delegate
 
+    @property
+    def is_dcr_bridge(self) -> bool:
+        """True when this client-forwarded-token server serves the gateway-hosted DCR front door
+        (gateway-self protected-resource and authorization-server metadata plus the register,
+        authorize, and token relays) instead of relaying the upstream's own OAuth discovery
+        verbatim. ``dcr_bridge`` is rejected on every other auth type at create, update, and
+        config load, so the mode gate here only defends rows edited outside those paths."""
+        return bool(self.dcr_bridge) and (self.is_true_passthrough or self.is_oauth_delegate)
+
     @property
     def requires_per_user_auth(self) -> bool:
         """

litellm/types/utils.py

@@ -2533,9 +2533,10 @@ class StandardLoggingMCPToolCall(TypedDict, total=False):
 
     mcp_server_resource: Optional[str]
     """
-    The upstream MCP server resource identifier (scheme + host + path) the tool call was
-    forwarded to. Redacted for logging: userinfo, query string, and fragment are stripped so an
-    upstream URL carrying an embedded token or secret query parameter never reaches log metadata.
+    The origin (scheme + host + port) of the upstream MCP server the tool call was forwarded
+    to. Redacted for logging: userinfo, the path, the query string, and the fragment are all
+    stripped, because hosted MCP servers routinely embed the credential in the URL path and
+    this value is readable by callers via request logs.
     Records which upstream received a relayed request; never a credential.
     """
 
@@ -2697,7 +2698,7 @@ class StandardLoggingGuardrailInformation(TypedDict, total=False):
     guardrail_name: Optional[str]
     guardrail_provider: Optional[str]
     guardrail_mode: Optional[Union[GuardrailEventHooks, List[GuardrailEventHooks], GuardrailMode]]
-    guardrail_request: Optional[dict]
+    guardrail_request: Optional[Union[str, dict]]
     guardrail_response: Optional[Union[dict, str, List[dict]]]
     guardrail_status: GuardrailStatus
     start_time: Optional[float]
@@ -2728,10 +2729,10 @@ class StandardLoggingGuardrailInformation(TypedDict, total=False):
     confidence_score: Optional[float]
     """For LLM-judge guardrails: confidence score 0.0-1.0"""
 
-    classification: Optional[dict]
+    classification: Optional[Union[str, dict]]
     """For LLM-judge guardrails: structured classification output"""
 
-    match_details: Optional[List[dict]]
+    match_details: Optional[Union[str, List[dict]]]
     """Detailed match information for each detected pattern"""
 
     patterns_checked: Optional[int]
@@ -3397,6 +3398,7 @@ class LlmProviders(str, Enum):
     LIBERTAI = "libertai"
     PINSTRIPES = "pinstripes"
     DARKBLOOM = "darkbloom"
+    META = "meta"
     LITELLM_AGENT = "litellm_agent"
     CURSOR = "cursor"
     BEDROCK_MANTLE = "bedrock_mantle"

litellm/utils.py

@@ -8028,6 +8028,16 @@ class ProviderConfigManager:
                 )
 
                 return GithubCopilotAnthropicMessagesConfig()
+
+        from litellm.llms.openai_like.json_loader import JSONProviderRegistry
+
+        json_provider = JSONProviderRegistry.get(provider.value)
+        if json_provider is not None and "/v1/messages" in json_provider.supported_endpoints:
+            from litellm.llms.openai_like.messages.transformation import (
+                JSONProviderAnthropicMessagesConfig,
+            )
+
+            return JSONProviderAnthropicMessagesConfig(json_provider)
         return None
 
     @staticmethod

model_prices_and_context_window.json

@@ -6012,6 +6012,522 @@
         "supports_vision": true,
         "supports_web_search": true
     },
+    "azure/gpt-5.6": {
+        "cache_read_input_token_cost": 5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1e-06,
+        "cache_read_input_token_cost_priority": 1e-06,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 2e-06,
+        "input_cost_per_token": 5e-06,
+        "input_cost_per_token_above_272k_tokens": 1e-05,
+        "input_cost_per_token_priority": 1e-05,
+        "input_cost_per_token_above_272k_tokens_priority": 2e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.5e-05,
+        "output_cost_per_token_priority": 6e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 9e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/gpt-5.6-sol": {
+        "cache_read_input_token_cost": 5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1e-06,
+        "cache_read_input_token_cost_priority": 1e-06,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 2e-06,
+        "input_cost_per_token": 5e-06,
+        "input_cost_per_token_above_272k_tokens": 1e-05,
+        "input_cost_per_token_priority": 1e-05,
+        "input_cost_per_token_above_272k_tokens_priority": 2e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.5e-05,
+        "output_cost_per_token_priority": 6e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 9e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/gpt-5.6-terra": {
+        "cache_read_input_token_cost": 2.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 5e-07,
+        "cache_read_input_token_cost_priority": 5e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 1e-06,
+        "input_cost_per_token": 2.5e-06,
+        "input_cost_per_token_above_272k_tokens": 5e-06,
+        "input_cost_per_token_priority": 5e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 1e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.5e-05,
+        "output_cost_per_token_above_272k_tokens": 2.25e-05,
+        "output_cost_per_token_priority": 3e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 4.5e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/gpt-5.6-luna": {
+        "cache_read_input_token_cost": 1e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 2e-07,
+        "cache_read_input_token_cost_priority": 2e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 4e-07,
+        "input_cost_per_token": 1e-06,
+        "input_cost_per_token_above_272k_tokens": 2e-06,
+        "input_cost_per_token_priority": 2e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 4e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 6e-06,
+        "output_cost_per_token_above_272k_tokens": 9e-06,
+        "output_cost_per_token_priority": 1.2e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 1.8e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/us/gpt-5.6": {
+        "cache_read_input_token_cost": 5.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1.1e-06,
+        "cache_read_input_token_cost_priority": 1.375e-06,
+        "input_cost_per_token": 5.5e-06,
+        "input_cost_per_token_above_272k_tokens": 1.1e-05,
+        "input_cost_per_token_priority": 1.375e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.95e-05,
+        "output_cost_per_token_priority": 8.25e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/us/gpt-5.6-sol": {
+        "cache_read_input_token_cost": 5.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1.1e-06,
+        "cache_read_input_token_cost_priority": 1.375e-06,
+        "input_cost_per_token": 5.5e-06,
+        "input_cost_per_token_above_272k_tokens": 1.1e-05,
+        "input_cost_per_token_priority": 1.375e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.95e-05,
+        "output_cost_per_token_priority": 8.25e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/us/gpt-5.6-terra": {
+        "cache_read_input_token_cost": 2.75e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 5.5e-07,
+        "cache_read_input_token_cost_priority": 6.875e-07,
+        "input_cost_per_token": 2.75e-06,
+        "input_cost_per_token_above_272k_tokens": 5.5e-06,
+        "input_cost_per_token_priority": 6.875e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.65e-05,
+        "output_cost_per_token_above_272k_tokens": 2.475e-05,
+        "output_cost_per_token_priority": 4.125e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/us/gpt-5.6-luna": {
+        "cache_read_input_token_cost": 1.1e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 2.2e-07,
+        "cache_read_input_token_cost_priority": 2.75e-07,
+        "input_cost_per_token": 1.1e-06,
+        "input_cost_per_token_above_272k_tokens": 2.2e-06,
+        "input_cost_per_token_priority": 2.75e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 6.6e-06,
+        "output_cost_per_token_above_272k_tokens": 9.9e-06,
+        "output_cost_per_token_priority": 1.65e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/eu/gpt-5.6": {
+        "cache_read_input_token_cost": 5.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1.1e-06,
+        "cache_read_input_token_cost_priority": 1.375e-06,
+        "input_cost_per_token": 5.5e-06,
+        "input_cost_per_token_above_272k_tokens": 1.1e-05,
+        "input_cost_per_token_priority": 1.375e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.95e-05,
+        "output_cost_per_token_priority": 8.25e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/eu/gpt-5.6-sol": {
+        "cache_read_input_token_cost": 5.5e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 1.1e-06,
+        "cache_read_input_token_cost_priority": 1.375e-06,
+        "input_cost_per_token": 5.5e-06,
+        "input_cost_per_token_above_272k_tokens": 1.1e-05,
+        "input_cost_per_token_priority": 1.375e-05,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.3e-05,
+        "output_cost_per_token_above_272k_tokens": 4.95e-05,
+        "output_cost_per_token_priority": 8.25e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/eu/gpt-5.6-terra": {
+        "cache_read_input_token_cost": 2.75e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 5.5e-07,
+        "cache_read_input_token_cost_priority": 6.875e-07,
+        "input_cost_per_token": 2.75e-06,
+        "input_cost_per_token_above_272k_tokens": 5.5e-06,
+        "input_cost_per_token_priority": 6.875e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.65e-05,
+        "output_cost_per_token_above_272k_tokens": 2.475e-05,
+        "output_cost_per_token_priority": 4.125e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
+    "azure/eu/gpt-5.6-luna": {
+        "cache_read_input_token_cost": 1.1e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 2.2e-07,
+        "cache_read_input_token_cost_priority": 2.75e-07,
+        "input_cost_per_token": 1.1e-06,
+        "input_cost_per_token_above_272k_tokens": 2.2e-06,
+        "input_cost_per_token_priority": 2.75e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 6.6e-06,
+        "output_cost_per_token_above_272k_tokens": 9.9e-06,
+        "output_cost_per_token_priority": 1.65e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_none_reasoning_effort": true,
+        "supports_xhigh_reasoning_effort": true,
+        "supports_minimal_reasoning_effort": false
+    },
     "azure/gpt-5.5": {
         "cache_read_input_token_cost": 5e-07,
         "cache_read_input_token_cost_above_272k_tokens": 1e-06,
@@ -25659,6 +26175,42 @@
         "supports_function_calling": true,
         "supports_tool_choice": false
     },
+    "meta/muse-spark-1.1": {
+        "cache_read_input_token_cost": 1.5e-07,
+        "input_cost_per_token": 1.25e-06,
+        "litellm_provider": "meta",
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 131072,
+        "max_tokens": 131072,
+        "mode": "chat",
+        "output_cost_per_token": 4.25e-06,
+        "source": "https://dev.meta.ai/docs/getting-started/pricing-rate-limits",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/responses",
+            "/v1/messages"
+        ],
+        "supported_modalities": [
+            "text",
+            "image",
+            "video"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_minimal_reasoning_effort": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_xhigh_reasoning_effort": true
+    },
     "meta_llama/Llama-3.3-70B-Instruct": {
         "litellm_provider": "meta_llama",
         "max_input_tokens": 128000,

provider_endpoints_support.json

@@ -1984,6 +1984,23 @@
         "interactions": true
       }
     },
+    "meta": {
+      "display_name": "Meta Model API (`meta`)",
+      "url": "https://docs.litellm.ai/docs/providers/meta",
+      "endpoints": {
+        "chat_completions": true,
+        "messages": true,
+        "responses": true,
+        "embeddings": false,
+        "image_generations": false,
+        "audio_transcriptions": false,
+        "audio_speech": false,
+        "moderations": false,
+        "batches": false,
+        "rerank": false,
+        "a2a": false
+      }
+    },
     "pinstripes": {
       "display_name": "Pinstripes (`pinstripes`)",
       "url": "https://docs.litellm.ai/docs/providers/pinstripes",

schema.prisma

@@ -339,6 +339,7 @@ model LiteLLM_MCPServerTable {
   available_on_public_internet Boolean @default(true)
   delegate_auth_to_upstream Boolean @default(false)
   oauth_passthrough Boolean @default(false)
+  dcr_bridge Boolean?
   is_byok               Boolean  @default(false)
   byok_description      String[] @default([])
   byok_api_key_help_url String?

tests/code_coverage_tests/recursive_detector.py

@@ -36,6 +36,7 @@ IGNORE_FUNCTIONS = [
     "_collect_argument_paths",  # max depth set.
     "_split_text",  # max depth set.
     "_mask_sequence",  # max depth set.
+    "_walk_payload",  # max depth set (DEFAULT_MAX_RECURSE_DEPTH_SENSITIVE_DATA_MASKER).
     "_delete_nested_value_custom",  # max depth set (bounded by number of path segments).
     "filter_exceptions_from_params",  # max depth set (default 20) to prevent infinite recursion.
     "__getattr__",  # lazy loading pattern in litellm/__init__.py with proper caching to prevent infinite recursion.

tests/e2e/batches/COVERAGE.md

@@ -16,8 +16,8 @@ failures are hard test failures (see `tests/e2e/CLAUDE.md`).
 |-----------|--------|----------|--------|------|--------------|
 | OpenAI    | yes | yes | yes | yes | OpenAI Files |
 | Azure     | yes | yes | yes | yes | Azure Files |
-| Vertex AI | yes | yes | yes | yes | GCS bucket (`GCS_BUCKET_NAME` via files_settings) |
-| Bedrock   | yes | yes | no (limited upstream) | no | S3 bucket (`AWS_BATCH_S3_BUCKET` + `AWS_BATCH_ROLE_ARN` on model) |
+| Vertex AI | yes | yes | yes | yes | GCS (`gcs_bucket_name` / `GCS_BUCKET_NAME` on model) |
+| Bedrock   | yes (unified only) | yes | no (limited upstream) | no | S3 (`s3_bucket_name` + `aws_*` + `AWS_BATCH_ROLE_ARN` on model) |
 
 Bedrock cancel is unreliable upstream and list is unsupported, so both are gated off
 (`can_cancel=False`, `can_list=False`) when that provider is enabled in the matrix.

tests/e2e/batches/capabilities.py

@@ -1,20 +1,22 @@
-"""The declarative provider x routing-scenario matrix the lifecycle test runs.
-
-One Capability per supported (provider, scenario) pair, so the parametrized test
-has no dead/skipped cells. `provider` is litellm's custom_llm_provider, used to
-route provider-fallback calls to /{provider}/v1/... and to assert the raw batch id
-shape (the only scenario whose id is not re-encoded by the proxy). Operations that
-a provider does not support (Bedrock: no cancel, no list) are gated per row.
-"""
+"""Provider x routing-scenario matrix for the batches lifecycle e2e."""
 
 from __future__ import annotations
 
 import base64
+import os
 from dataclasses import dataclass
 from typing import Literal
 
 from models import LiteLLMParamsBody
 
+
+def _env_ref(*names: str) -> str:
+    for name in names:
+        value = os.environ.get(name)
+        if value is not None and value.strip() != "":
+            return f"os.environ/{name}"
+    return f"os.environ/{names[0]}"
+
 Scenario = Literal["encoded", "unified", "model_param", "provider_fallback"]
 
 IdShape = Literal["managed", "model_encoded", "raw"]
@@ -44,10 +46,10 @@ class Provider:
                 )
             case "azure":
                 return LiteLLMParamsBody(
-                    model="azure/gpt-4.1-mini-batch",
+                    model="azure/gpt-5.4-mini-batch",
                     api_base="os.environ/AZURE_API_BASE",
                     api_key="os.environ/AZURE_API_KEY",
-                    api_version="2024-07-01-preview",
+                    api_version="2025-04-01-preview",
                 )
             case "vertex_ai":
                 return LiteLLMParamsBody(
@@ -55,14 +57,19 @@ class Provider:
                     vertex_project="os.environ/VERTEXAI_PROJECT",
                     vertex_location="us-central1",
                     vertex_credentials="os.environ/VERTEXAI_CREDENTIALS",
+                    gcs_bucket_name="os.environ/GCS_BUCKET_NAME",
+                    bucket_name="os.environ/GCS_BUCKET_NAME",
                 )
             case "bedrock":
                 return LiteLLMParamsBody(
                     model="bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0",
+                    aws_access_key_id="os.environ/AWS_ACCESS_KEY_ID",
+                    aws_secret_access_key="os.environ/AWS_SECRET_ACCESS_KEY",
+                    aws_region_name="os.environ/AWS_REGION",
+                    s3_region_name="os.environ/AWS_REGION",
+                    s3_bucket_name=_env_ref("AWS_BATCH_S3_BUCKET", "AWS_S3_BUCKET_NAME"),
                     s3_access_key_id="os.environ/AWS_ACCESS_KEY_ID",
                     s3_secret_access_key="os.environ/AWS_SECRET_ACCESS_KEY",
-                    s3_region_name="os.environ/AWS_REGION",
-                    s3_bucket_name="os.environ/AWS_BATCH_S3_BUCKET",
                     aws_batch_role_arn="os.environ/AWS_BATCH_ROLE_ARN",
                 )
             case _:
@@ -84,20 +91,12 @@ class Capability:
 
     @property
     def jsonl_model(self) -> str:
-        """Model name embedded in the uploaded JSONL ``body.model``.
-
-        Only the unified upload path rewrites JSONL on upload
-        (``target_model_names`` → ``llm_router.acreate_file`` →
-        ``replace_model_in_jsonl``), so that scenario can use the LiteLLM alias
-        and rely on the proxy to swap it to the deployment model. Every other
-        scenario uploads raw JSONL with no rewrite, so the provider's real
-        deployment name is required or create fails upstream validation."""
         return self.model if self.scenario == "unified" else self.raw_model
 
 
 PROVIDERS: tuple[Provider, ...] = (
     Provider("openai", "openai-batch", "gpt-4o-mini", can_cancel=True, can_list=True),
-    Provider("azure", "azure-batch", "gpt-4.1-mini-batch", can_cancel=True, can_list=True),
+    Provider("azure", "azure-batch", "gpt-5.4-mini-batch", can_cancel=True, can_list=True),
     Provider(
         "vertex_ai", "vertex-batch", "gemini-2.5-flash", can_cancel=True, can_list=True
     ),
@@ -110,7 +109,7 @@ PROVIDERS: tuple[Provider, ...] = (
     ),
 )
 
-BEDROCK_SCENARIOS: tuple[Scenario, ...] = ("encoded", "unified")
+BEDROCK_SCENARIOS: tuple[Scenario, ...] = ("unified",)
 
 
 def scenarios_for_provider(provider: Provider) -> tuple[Scenario, ...]:
@@ -127,8 +126,6 @@ CAPABILITIES: tuple[Capability, ...] = tuple(
 
 
 def raw_id_matches_provider(provider: str, batch_id: str) -> bool:
-    """The provider-fallback path returns the provider's native batch id (unencoded),
-    so its shape discriminates which provider actually handled the batch."""
     if provider in ("openai", "azure"):
         return batch_id.startswith("batch")
     if provider == "vertex_ai":
@@ -166,12 +163,10 @@ def _b64_decode(value: str) -> str:
 
 
 def is_managed_id(id_str: str) -> bool:
-    """A litellm managed unified file/batch id base64-decodes to a litellm_proxy marker."""
     return _b64_decode(id_str).startswith("litellm_proxy")
 
 
 def is_model_encoded_id(id_str: str) -> bool:
-    """A model-encoded id keeps the provider prefix and base64-encodes litellm:<id>;model,<m>."""
     for prefix in ("file-", "batch_"):
         if id_str.startswith(prefix):
             decoded = _b64_decode(id_str[len(prefix) :])

tests/e2e/batches/test_batches_e2e.py

@@ -142,10 +142,12 @@ def quietly(action: Callable[[], object]) -> Callable[[], None]:
     return run
 
 
-def assert_file_object(file: FileObject) -> None:
+def assert_file_object(file: FileObject, *, provider: str) -> None:
     assert file.object == "file", f"file.object={file.object!r}"
     assert file.purpose == "batch", f"file.purpose={file.purpose!r}"
-    assert file.bytes is not None and file.bytes > 0, f"file.bytes={file.bytes!r}"
+    assert file.bytes is not None, f"file.bytes={file.bytes!r}"
+    if provider != "bedrock":
+        assert file.bytes > 0, f"file.bytes={file.bytes!r}"
     assert file.status, "file.status missing"
     assert (
         file.created_at is not None and file.created_at > 0
@@ -179,7 +181,7 @@ def test_batch_lifecycle(
     resources.defer(
         quietly(lambda: client.delete_file(file.id, key=key, provider=provider))
     )
-    assert_file_object(file)
+    assert_file_object(file, provider=cap.provider)
     assert matches_id_shape(
         FILE_ID_SHAPE[cap.scenario], file.id
     ), f"{cap.id}: file id {file.id!r} is not a {FILE_ID_SHAPE[cap.scenario]} id"
@@ -234,10 +236,31 @@ def test_batch_lifecycle(
         )
 
     if cap.can_list:
-        listed = unwrap(client.list_batches(key=key, provider=provider))
+        list_result = client.list_batches(key=key, provider=provider)
+        managed_filter_unsupported = False
+        match list_result:
+            case UnknownApiError(body=body) if (
+                "Filtering by 'provider' is not supported when using managed batches" in body
+            ):
+                managed_filter_unsupported = True
+                listed = unwrap(client.list_batches(key=key, provider=None))
+            case _:
+                listed = unwrap(list_result)
         if listed.object is not None:
             assert listed.object == "list", f"list envelope object={listed.object!r}"
         match = next((b for b in listed.data if b.id == batch.id), None)
+        if (
+            match is None
+            and managed_filter_unsupported
+            and cap.scenario == "provider_fallback"
+        ):
+            # provider_fallback keeps the provider's raw batch id (not re-encoded
+            # into a managed/proxy id). When the gateway rejects provider-scoped
+            # list, the only available list is the unfiltered managed view, which
+            # does not index raw provider ids. Membership cannot be asserted here;
+            # create + retrieve (and raw_id_matches_provider above) already pin
+            # routing for this scenario.
+            return
         assert match is not None, "created batch absent from list"
         assert match.object == "batch"
 
@@ -289,7 +312,7 @@ def test_file_upload_and_delete_outputs(
             key=key,
         )
     )
-    assert_file_object(file)
+    assert_file_object(file, provider="openai")
 
     deleted = unwrap(client.delete_file(file.id, key=key))
     assert deleted.id, "delete response has no id"

tests/e2e/budgets/budget_client.py

@@ -10,12 +10,13 @@ and response models are co-located here because only this suite uses them.
 
 from __future__ import annotations
 
+import time
 from dataclasses import dataclass
 
 from pydantic import AliasPath, BaseModel, Field, RootModel
 
 from e2e_gateway import Gateway, build_gateway
-from e2e_http import NoBody, StreamingResponse, Success, unwrap
+from e2e_http import NoBody, Result, StreamingResponse, Success, unwrap
 from models import (
     AnthropicMessagesBody,
     BudgetWindow,
@@ -26,6 +27,9 @@ from models import (
     ModelBudgetEntry,
 )
 
+_TEAM_READY_ATTEMPTS = 15
+_TEAM_READY_SLEEP_SECONDS = 0.4
+
 
 class UserNewBody(BaseModel):
     max_budget: float
@@ -299,7 +303,7 @@ class BudgetClient:
         organization_id: str | None = None,
         budget_limits: list[BudgetWindow] | None = None,
     ) -> str:
-        return unwrap(
+        team_id = unwrap(
             self.gateway.transport.post(
                 "/team/new",
                 headers=self.gateway.transport.master,
@@ -312,6 +316,8 @@ class BudgetClient:
                 response_type=TeamNewResponse,
             )
         ).team_id
+        self._wait_for_team(team_id)
+        return team_id
 
     def delete_team(self, team_id: str) -> None:
         _ = self.gateway.transport.post(
@@ -321,17 +327,43 @@ class BudgetClient:
             response_type=NoBody,
         )
 
+    def _wait_for_team(self, team_id: str) -> None:
+        last: Result[TeamInfoResponse] | None = None
+        for _ in range(_TEAM_READY_ATTEMPTS):
+            last = self.gateway.transport.get(
+                "/team/info",
+                headers=self.gateway.transport.master,
+                params=TeamInfoParams(team_id=team_id),
+                response_type=TeamInfoResponse,
+            )
+            match last:
+                case Success():
+                    return
+                case _:
+                    time.sleep(_TEAM_READY_SLEEP_SECONDS)
+        assert last is not None
+        _ = unwrap(last)
+
     def add_team_member(self, team_id: str, user_id: str, *, max_budget_in_team: float | None = None) -> None:
-        resp = self.gateway.transport.send(
-            "/team/member_add",
-            headers=self.gateway.transport.master,
-            json=TeamMemberAddBody(
-                team_id=team_id,
-                member=TeamMember(role="user", user_id=user_id),
-                max_budget_in_team=max_budget_in_team,
-            ),
-        )
-        assert resp.ok, resp.body
+        last_body = ""
+        for attempt in range(_TEAM_READY_ATTEMPTS):
+            resp = self.gateway.transport.send(
+                "/team/member_add",
+                headers=self.gateway.transport.master,
+                json=TeamMemberAddBody(
+                    team_id=team_id,
+                    member=TeamMember(role="user", user_id=user_id),
+                    max_budget_in_team=max_budget_in_team,
+                ),
+            )
+            if resp.ok:
+                return
+            last_body = resp.body
+            if "doesn't exist" in resp.body and attempt + 1 < _TEAM_READY_ATTEMPTS:
+                time.sleep(_TEAM_READY_SLEEP_SECONDS)
+                continue
+            break
+        assert False, last_body
 
     def update_team_member(
         self,

tests/e2e/budgets/test_budget_reset_advances_e2e.py

@@ -0,0 +1,228 @@
+"""Live e2e: regression guards for #25109 (budget resets stopped working).
+
+The existing test_budget_reset_e2e.py / test_multi_window_budget_e2e.py prove a
+blocked key flows again after its window. #25109 stored multi-budget-window data
+in nullable JSON columns and filtered eligible rows with a `not: None`-style Prisma
+filter that misbehaves on a nullable JSON column, so due rows were either skipped
+(budget_reset_at stayed pinned, spend never cleared) or the reset path errored
+(a non-budget 5xx leaked to callers). These tests assert the precise invariants
+that bug broke, built up START-SLOW from scheduling -> enforcement -> the reset
+strictly advancing -> the JSON-backed multi-window / team-member edges -> the
+error path. They EXTEND the happy-path modules rather than duplicate them: each
+asserts a delta (before<after timestamp, independent windows, block-not-error)
+that the happy-path "calls flow again" check alone does not pin down.
+"""
+
+import time
+from datetime import datetime
+
+import pytest
+
+from budget_client import BudgetClient, is_budget_block
+from e2e_config import unique_marker
+from e2e_http import require_successful_call
+from lifecycle import ResourceManager
+from models import BudgetWindow
+
+pytestmark = pytest.mark.e2e
+
+WINDOW_SECONDS = 30
+RESET_DEADLINE_SECONDS = 150
+TINY_CAP = 3e-6
+
+
+def _call(client: BudgetClient, key: str):
+    return client.chat(key, "claude-haiku-4-5", f"advance {unique_marker()}", max_tokens=16)
+
+
+def _as_datetime(value: str) -> datetime:
+    return datetime.fromisoformat(value.replace("Z", "+00:00"))
+
+
+def _drive_to_block(client: BudgetClient, key: str) -> None:
+    """Spend until the cap blocks; fails loudly if enforcement never trips."""
+    for _ in range(20):
+        result = _call(client, key)
+        if is_budget_block(result):
+            return
+        require_successful_call(result)
+        time.sleep(2)
+    pytest.fail("budget never enforced before block")
+
+
+# ---- Rung 1: scheduling exists at creation -----------------------------------
+
+
+def test_key_with_budget_duration_schedules_reset_at_creation(
+    client: BudgetClient, resources: ResourceManager
+) -> None:
+    """Baseline: a key created with a budget_duration has budget_reset_at populated
+    immediately. The reset job can only advance a timestamp that was scheduled in
+    the first place; everything below depends on this."""
+    key = client.generate_key(max_budget=TINY_CAP, budget_duration=f"{WINDOW_SECONDS}s")
+    resources.defer(lambda: client.delete_key(key))
+
+    info = client.gateway.key_info(key)
+    assert info.budget_reset_at is not None, "budget_duration set no budget_reset_at"
+    assert _as_datetime(info.budget_reset_at) > _as_datetime("1970-01-01T00:00:00Z")
+
+
+# ---- Rung 2: enforcement trips at the cap ------------------------------------
+
+
+def test_key_spend_blocks_at_cap(client: BudgetClient, resources: ResourceManager) -> None:
+    """Sanity that the tiny cap is enforced before we test that it resets: spend
+    accrues across calls and eventually returns budget_exceeded, never a 5xx."""
+    key = client.generate_key(max_budget=TINY_CAP, budget_duration=f"{WINDOW_SECONDS}s")
+    resources.defer(lambda: client.delete_key(key))
+
+    # _drive_to_block is the enforcement proof: it fails unless a budget_exceeded
+    # block follows successful (non-5xx) calls. key_info.spend is deliberately not
+    # asserted - it is the DB-persisted field that flushes ~60s later
+    # (proxy_batch_write_at), so reading it right after the block races to 0.0.
+    _drive_to_block(client, key)
+
+
+# ---- Rung 3: the core regression - reset_at strictly advances + spend zeroes --
+
+
+def test_key_budget_reset_at_advances_after_window(
+    client: BudgetClient, resources: ResourceManager
+) -> None:
+    """The core #25109 guard: after the window elapses the reset job must move
+    budget_reset_at strictly forward AND zero key.spend. The broken nullable-JSON
+    filter left eligible rows untouched, so the timestamp stayed pinned and spend
+    never cleared. Asserting before<after (not merely "a call succeeded") kills a
+    mutation that no-ops the reset while leaving enforcement intact."""
+    key = client.generate_key(max_budget=TINY_CAP, budget_duration=f"{WINDOW_SECONDS}s")
+    resources.defer(lambda: client.delete_key(key))
+
+    before_raw = client.gateway.key_info(key).budget_reset_at
+    assert before_raw is not None, "no budget_reset_at scheduled at creation"
+    before = _as_datetime(before_raw)
+
+    _drive_to_block(client, key)
+
+    deadline = time.monotonic() + RESET_DEADLINE_SECONDS
+    while time.monotonic() < deadline:
+        time.sleep(5)
+        result = _call(client, key)
+        if not result.ok:
+            assert is_budget_block(result), f"non-budget error during reset wait: {result.body[:200]}"
+            continue
+        info = client.gateway.key_info(key)
+        assert info.budget_reset_at is not None, "budget_reset_at cleared by reset"
+        assert _as_datetime(info.budget_reset_at) > before, (
+            "budget_reset_at did not advance past the pre-reset value"
+        )
+        assert (info.spend or 0.0) < TINY_CAP, f"spend not cleared after reset: {info.spend}"
+        return
+    pytest.fail(f"key budget never reset within {RESET_DEADLINE_SECONDS}s")
+
+
+# ---- Rung 4: multi-window - tight window resets, roomy window keeps spend -----
+
+
+def test_multi_window_key_resets_each_window_independently(
+    client: BudgetClient, resources: ResourceManager
+) -> None:
+    """The JSON-backed path #25109 specifically touched. A tight 30s window and a
+    roomy 1m window: the tight window must reset on its own boundary while the roomy
+    window keeps its accumulated spend (independent per-window reset). The
+    nullable-JSON filter bug skipped these JSON-backed rows entirely, so the tight
+    window never came back; a job that ERRORS on the JSON column would surface here
+    as a non-budget 5xx, which we reject throughout the wait."""
+    key = client.generate_key(
+        budget_limits=[
+            BudgetWindow(budget_duration=f"{WINDOW_SECONDS}s", max_budget=TINY_CAP),
+            BudgetWindow(budget_duration="1m", max_budget=1.0),
+        ]
+    )
+    resources.defer(lambda: client.delete_key(key))
+
+    start = time.monotonic()
+    _drive_to_block(client, key)
+    spend_at_block = client.gateway.key_info(key).spend or 0.0
+
+    deadline = time.monotonic() + RESET_DEADLINE_SECONDS
+    while time.monotonic() < deadline:
+        time.sleep(5)
+        result = _call(client, key)
+        if result.ok:
+            elapsed = time.monotonic() - start
+            assert elapsed < WINDOW_SECONDS + 90, (
+                f"tight window reset took {elapsed:.0f}s - too long for {WINDOW_SECONDS}s"
+            )
+            assert (client.gateway.key_info(key).spend or 0.0) >= spend_at_block, (
+                "roomy window spend was wiped when only the tight window should reset"
+            )
+            return
+        assert is_budget_block(result), f"non-budget error during reset wait: {result.body[:200]}"
+    pytest.fail(f"tight window never reset within {RESET_DEADLINE_SECONDS}s")
+
+
+# ---- Rung 5: team-member window advances (JSON-backed per-team budget) --------
+
+
+def test_team_member_budget_reset_at_advances(
+    client: BudgetClient, resources: ResourceManager
+) -> None:
+    """Per-team member windows are also JSON-backed. member_budget_reset_at must
+    advance after the window; the explicit before<after assertion is the #25109
+    regression guard (the existing reset test only checks "it eventually moved",
+    this pins it strictly past the value recorded before the window)."""
+    team_id = client.create_team(alias=f"e2e-member-advance-{unique_marker()}", max_budget=100.0)
+    resources.defer(lambda: client.delete_team(team_id))
+    user_id = client.create_user(max_budget=100.0)
+    resources.defer(lambda: client.delete_user(user_id))
+
+    client.add_team_member(team_id, user_id, max_budget_in_team=1.0)
+    client.update_team_member(
+        team_id, user_id, max_budget_in_team=1.0, budget_duration=f"{WINDOW_SECONDS}s"
+    )
+
+    before_raw = client.member_budget_reset_at(team_id, user_id)
+    assert before_raw, "updating the member with a budget_duration set no budget_reset_at"
+    before = _as_datetime(before_raw)
+
+    key = client.generate_key(team_id=team_id, user_id=user_id)
+    resources.defer(lambda: client.delete_key(key))
+    require_successful_call(_call(client, key))
+
+    deadline = time.monotonic() + RESET_DEADLINE_SECONDS
+    while time.monotonic() < deadline:
+        time.sleep(5)
+        current = client.member_budget_reset_at(team_id, user_id)
+        if current and _as_datetime(current) > before:
+            return
+    pytest.fail(f"member budget_reset_at never advanced past {before.isoformat()} in {RESET_DEADLINE_SECONDS}s")
+
+
+# ---- Rung 6: error-path edge - resets surface as blocks, never 5xx -----------
+
+
+def test_reset_wait_never_yields_non_budget_error(
+    client: BudgetClient, resources: ResourceManager
+) -> None:
+    """The other #25109 failure mode: a reset job that ERRORS on the nullable-JSON
+    column surfaces to the caller as a non-budget 5xx. Across the whole reset wait
+    every non-ok response must be a budget block (is_budget_block) and never a
+    server error; this guards the error path independently of whether the reset
+    eventually fires."""
+    key = client.generate_key(max_budget=TINY_CAP, budget_duration=f"{WINDOW_SECONDS}s")
+    resources.defer(lambda: client.delete_key(key))
+
+    _drive_to_block(client, key)
+
+    saw_reset = False
+    deadline = time.monotonic() + RESET_DEADLINE_SECONDS
+    while time.monotonic() < deadline:
+        time.sleep(5)
+        result = _call(client, key)
+        if result.ok:
+            saw_reset = True
+            break
+        assert is_budget_block(result), (
+            f"reset wait yielded a non-budget error (likely a JSON-column reset crash): {result.body[:200]}"
+        )
+    assert saw_reset, f"key budget never reset within {RESET_DEADLINE_SECONDS}s"

tests/e2e/budgets/test_spend_counter_reseed_e2e.py

@@ -64,25 +64,47 @@ def _redis() -> "redis.Redis[str] | RedisCluster[str]":
     )
 
 
+def _parse_counter(raw: object) -> float | None:
+    if raw is None:
+        return None
+    if isinstance(raw, (int, float)):
+        return float(raw)
+    text = str(raw).strip()
+    if not text:
+        return None
+    try:
+        return float(text)
+    except ValueError:
+        try:
+            import json
+
+            return float(json.loads(text))
+        except Exception:
+            return None
+
+
 def _spend_counter(rds: "redis.Redis[str] | RedisCluster[str]", key: str) -> float | None:
-    """The shared spend counter for `key`, or None if it is cold. A cluster client
-    can't run a keyspace SCAN that spans shards, so read the key directly - the stage
-    gateway sets no cache namespace, so the key is the bare ``spend:key:{sha256(key)}``.
-    A standalone client matches by suffix, so the local cache namespace (litellm.caching)
-    need not be hard-coded here."""
+    """The shared spend counter for `key`, or None if it is cold.
+
+    The gateway keys counters as ``spend:key:{sha256(raw_sk)}``, optionally under a
+    redis namespace prefix. Cluster mode cannot SCAN all shards, so try the bare key
+    and a few common namespaces; standalone redis uses a suffix SCAN.
+    """
     from redis.cluster import RedisCluster
 
     digest = hashlib.sha256(key.encode()).hexdigest()
     suffix = f"spend:key:{digest}"
     if isinstance(rds, RedisCluster):
-        raw = rds.get(suffix)
-        return float(raw) if raw is not None else None
+        for candidate in (suffix, f"litellm:{suffix}", f"litellm.caching:{suffix}"):
+            parsed = _parse_counter(rds.get(candidate))
+            if parsed is not None:
+                return parsed
+        return None
 
     matches = list(rds.scan_iter(match=f"*{suffix}"))
     if not matches:
         return None
-    raw = rds.get(matches[0])
-    return float(raw) if raw is not None else None
+    return _parse_counter(rds.get(matches[0]))
 
 
 def _chat(client: BudgetClient, key: str) -> StreamingResponse:
@@ -97,19 +119,6 @@ def _accumulate(client: BudgetClient, key: str, count: int) -> None:
         list(pool.map(one, range(count)))
 
 
-def _burst(client: BudgetClient, key: str, count: int) -> None:
-    """Fire `count` requests that start together, so multiple workers reseed the cold
-    counter concurrently rather than one warming it before the others arrive."""
-    barrier = Barrier(count)
-
-    def one(_: int) -> StreamingResponse:
-        barrier.wait()
-        return _chat(client, key)
-
-    with ThreadPoolExecutor(max_workers=count) as pool:
-        list(pool.map(one, range(count)))
-
-
 def test_cold_counter_reseed_keeps_counter_equal_to_db_spend(
     client: BudgetClient, resources: ResourceManager
 ) -> None:
@@ -132,10 +141,28 @@ def test_cold_counter_reseed_keeps_counter_equal_to_db_spend(
     db_spend = client.gateway.key_info(key).spend or 0.0
     assert db_spend > 0, f"no DB spend accumulated from real calls: {db_spend}"
 
-    _burst(client, key, BURST)
-    time.sleep(3)
+    burst_results = []
+    barrier = Barrier(BURST)
+
+    def one(_: int) -> StreamingResponse:
+        barrier.wait()
+        return _chat(client, key)
+
+    with ThreadPoolExecutor(max_workers=BURST) as pool:
+        burst_results = list(pool.map(one, range(BURST)))
+    assert all(r.ok for r in burst_results), (
+        "some burst calls failed; cannot exercise concurrent reseed. "
+        f"statuses={[r.status_code for r in burst_results]}"
+    )
+
+    counter: float | None = None
+    deadline = time.monotonic() + 15
+    while time.monotonic() < deadline:
+        counter = _spend_counter(rds, key)
+        if counter is not None:
+            break
+        time.sleep(0.5)
 
-    counter = _spend_counter(rds, key)
     assert counter is not None, "the burst did not reseed the cold counter"
     assert db_spend * 0.95 <= counter < db_spend * 1.7, (
         f"redis spend counter {counter} does not equal DB spend {db_spend} (expected ~equal "

tests/e2e/budgets/test_tag_budget_e2e.py

@@ -26,7 +26,7 @@ def _tagged_call(client: BudgetClient, key: str, tag: str):
         "claude-haiku-4-5",
         f"hi {unique_marker()}",
         tags=[tag],
-        max_tokens=16,
+        max_tokens=64,
     )
     if not result.ok and not is_budget_block(result):
         require_successful_call(result)
@@ -40,17 +40,20 @@ def test_tag_budget_blocks_tagged_requests(
     client.create_tag(budgeted_tag, max_budget=TINY_BUDGET)
     resources.defer(lambda: client.delete_tag(budgeted_tag))
 
-    # Requests under the budgeted tag get blocked once its spend is exceeded.
-    blocked = False
-    deadline = time.monotonic() + 60
-    while time.monotonic() < deadline:
-        if is_budget_block(_tagged_call(client, scoped_key, budgeted_tag)):
-            blocked = True
-            break
-        time.sleep(1)
+    first = _tagged_call(client, scoped_key, budgeted_tag)
+    if is_budget_block(first):
+        blocked = True
+    else:
+        require_successful_call(first)
+        blocked = False
+        deadline = time.monotonic() + 120
+        while time.monotonic() < deadline:
+            if is_budget_block(_tagged_call(client, scoped_key, budgeted_tag)):
+                blocked = True
+                break
+            time.sleep(1)
     assert blocked, f"tag budget for {budgeted_tag!r} never enforced"
 
-    # A request with an unbudgeted tag on the same key is unaffected.
     free_tag = f"e2e-free-tag-{unique_marker()}"
     other = _tagged_call(client, scoped_key, free_tag)
     assert not is_budget_block(other), (

tests/e2e/budgets/test_team_multi_window_budget_e2e.py

@@ -41,19 +41,19 @@ def test_team_short_window_blocks_then_resets(client: BudgetClient, resources: R
         ],
     )
     resources.defer(lambda: client.delete_team(team_id))
-    key = client.generate_key(team_id=team_id)
+    key = client.generate_key(team_id=team_id, models=["claude-haiku-4-5"])
     resources.defer(lambda: client.delete_key(key))
 
     # 1. exhaust the tight window -> litellm returns budget_exceeded
     start = time.monotonic()
     blocked = False
-    for _ in range(20):
+    for _ in range(30):
         result = _call(client, key)
         if is_budget_block(result):
             blocked = True
             break
         require_successful_call(result)
-        time.sleep(2)
+        time.sleep(1)
     assert blocked, f"team {WINDOW_SECONDS}s window never enforced"
 
     # 2. the window resets at the next wall-clock-aligned boundary (up to a window

tests/e2e/docker-compose.yml

@@ -6,6 +6,8 @@ configs:
         master_key: os.environ/LITELLM_MASTER_KEY
         database_url: os.environ/DATABASE_URL
         store_prompts_in_spend_logs: true
+        proxy_budget_rescheduler_min_time: 5
+        proxy_budget_rescheduler_max_time: 10
 
       litellm_settings:
         drop_params: true
@@ -29,6 +31,14 @@ configs:
         - custom_llm_provider: openai
           api_key: os.environ/OPENAI_API_KEY
 
+      files_settings:
+        - custom_llm_provider: openai
+          api_key: os.environ/OPENAI_API_KEY
+        - custom_llm_provider: azure
+          api_base: os.environ/AZURE_API_BASE
+          api_key: os.environ/AZURE_API_KEY
+          api_version: "2024-05-01-preview"
+
       model_list:
         - model_name: gpt-5.5
           litellm_params:
@@ -64,6 +74,19 @@ services:
       DATABASE_URL: postgresql://litellm:litellm@db:5432/litellm
       UI_USERNAME: admin
       UI_PASSWORD: sk-1234
+      AWS_S3_BUCKET_NAME: ${AWS_S3_BUCKET_NAME:-${AWS_BATCH_S3_BUCKET:-}}
+      AWS_BATCH_S3_BUCKET: ${AWS_BATCH_S3_BUCKET:-${AWS_S3_BUCKET_NAME:-}}
+      AWS_BATCH_ROLE_ARN: ${AWS_BATCH_ROLE_ARN:-}
+      AWS_ACCESS_KEY_ID: ${AWS_ACCESS_KEY_ID:-}
+      AWS_SECRET_ACCESS_KEY: ${AWS_SECRET_ACCESS_KEY:-}
+      AWS_REGION: ${AWS_REGION:-us-east-1}
+      GCS_BUCKET_NAME: ${GCS_BUCKET_NAME:-}
+      VERTEXAI_PROJECT: ${VERTEXAI_PROJECT:-}
+      VERTEXAI_CREDENTIALS: ${VERTEXAI_CREDENTIALS:-}
+      GOOGLE_APPLICATION_CREDENTIALS: ${GOOGLE_APPLICATION_CREDENTIALS:-}
+      MISTRAL_API_KEY: ${MISTRAL_API_KEY:-}
+      AZURE_API_BASE: ${AZURE_API_BASE:-}
+      AZURE_API_KEY: ${AZURE_API_KEY:-}
     ports:
       - "4000:4000"
     configs:

tests/e2e/llm_translation/realtime/REALTIME_COVERAGE_MATRIX.md

@@ -43,7 +43,7 @@ at call time. The provider table below is the source of truth; edit `PROVIDERS`
 | gemini | `gemini-realtime` | `gemini/gemini-3.1-flash-live-preview` |
 | vertex_ai | `vertex-realtime` | `vertex_ai/gemini-live-2.5-flash-preview-native-audio-09-2025` |
 
-Bedrock and xai (`xai/grok-4-1-fast-non-reasoning`) are supported by the proxy but
+Bedrock and xai (`xai/grok-4-1-fast`) are supported by the proxy but
 kept commented out in `PROVIDERS` until they pass end-to-end here; re-enable them by
 uncommenting their entry.
 

tests/e2e/llm_translation/realtime/realtime_client.py

@@ -95,7 +95,7 @@ PROVIDERS = (
     #     "xai",
     #     "xai-realtime",
     #     LiteLLMParamsBody(
-    #         model="xai/grok-4-1-fast-non-reasoning",
+    #         model="xai/grok-4-1-fast",
     #         api_key="os.environ/XAI_API_KEY",
     #     ),
     # ),  # TODO: Enable once xai Grok Voice realtime is passing end-to-end here

tests/e2e/llm_translation/realtime/test_realtime_pipecat_audio_e2e.py

@@ -38,6 +38,13 @@ pytestmark = pytest.mark.e2e
 
 pytest.importorskip("pipecat", reason="pipecat-ai not installed")
 
+try:
+    import nltk
+
+    nltk.data.find("tokenizers/punkt_tab")
+except LookupError:
+    pytest.skip("NLTK punkt_tab data is not installed", allow_module_level=True)
+
 from pipecat.adapters.schemas.function_schema import FunctionSchema  # noqa: E402
 from pipecat.adapters.schemas.tools_schema import ToolsSchema  # noqa: E402
 from pipecat.frames.frames import (  # noqa: E402
@@ -96,7 +103,7 @@ SERVER_VAD_SETTINGS = rt_events.SessionProperties(
             noise_reduction=rt_events.InputAudioNoiseReduction(type="near_field"),
             turn_detection=rt_events.TurnDetection(
                 type="server_vad",
-                threshold=0.8,
+                threshold=0.5,
                 prefix_padding_ms=300,
                 silence_duration_ms=700,
             ),

tests/e2e/llm_translation/test_cache_control.py

@@ -0,0 +1,163 @@
+"""Live e2e: provider-specific /chat/completions features take real effect.
+
+Each case asserts the feature actually happened, not just a 200. Coverage matrix
+(register-on-demand deployments, deleted on teardown):
+
+- Bedrock (anthropic claude-haiku-4-5): prompt caching. A large cacheable prefix
+  marked with ``cache_control`` is sent twice; the second call must report
+  cache-read usage tokens > 0. service_tier is out of scope for Bedrock; AWS
+  Bedrock does not expose an OpenAI-style request service tier, so that cell is
+  intentionally not covered here.
+- Vertex (gemini-2.5-flash): prompt caching via ``cache_control`` context
+  caching; the second identical call must report cached prompt tokens > 0.
+
+service_tier lives in test_provider_features_e2e.py.
+
+The provider-native cache_control request shape is not expressible with the
+shared ``ChatBody`` (whose content is a plain string), so the cacheable body is
+modelled locally with typed content blocks.
+"""
+
+from __future__ import annotations
+
+import time
+
+import pytest
+from pydantic import BaseModel
+
+from e2e_config import unique_marker
+from e2e_http import Result, unwrap
+from lifecycle import ResourceManager
+from models import ChatResponse, LiteLLMParamsBody, Usage
+from passthrough_client import PassthroughClient
+import os
+
+pytestmark = pytest.mark.e2e
+
+BEDROCK_MODEL = "bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0"
+VERTEX_MODEL = "vertex_ai/gemini-2.5-flash"
+
+
+class CacheControl(BaseModel):
+    type: str = "ephemeral"
+
+
+class TextBlock(BaseModel):
+    type: str = "text"
+    text: str
+    cache_control: CacheControl | None = None
+
+
+class RichMessage(BaseModel):
+    role: str
+    content: list[TextBlock]
+
+
+class CacheChatBody(BaseModel):
+    model: str
+    messages: list[RichMessage]
+    max_tokens: int = 64
+    cache: dict[str, bool] = {"no-cache": True}
+
+
+def _cacheable_prefix() -> str:
+    """A prefix long enough to clear provider minimum cacheable sizes (Haiku is
+    2048 tokens), unique per run so the first call writes and the second reads."""
+    marker = unique_marker()
+    body = " ".join(
+        f"Cacheable reference paragraph {index} for run {marker}." for index in range(600)
+    )
+    return f"{body}\nEnd of reference material {marker}."
+
+
+def _cached_read_tokens(usage: Usage | None) -> int:
+    """Cache-read tokens however the provider reports them: Anthropic-style
+    ``cache_read_input_tokens`` or OpenAI-style ``prompt_tokens_details.cached_tokens``."""
+    if usage is None:
+        return 0
+    if usage.cache_read_input_tokens:
+        return usage.cache_read_input_tokens
+    if usage.prompt_tokens_details and usage.prompt_tokens_details.cached_tokens:
+        return usage.prompt_tokens_details.cached_tokens
+    return 0
+
+
+def _cache_chat(
+    client: PassthroughClient, key: str, model: str, prefix: str
+) -> Result[ChatResponse]:
+    body = CacheChatBody(
+        model=model,
+        messages=[
+            RichMessage(
+                role="system",
+                content=[TextBlock(text=prefix, cache_control=CacheControl())],
+            ),
+            RichMessage(role="user", content=[TextBlock(text="Reply with one word.")]),
+        ],
+    )
+    return client.gateway.transport.post(
+        "/chat/completions",
+        headers=client.gateway.transport.bearer(key),
+        json=body,
+        response_type=ChatResponse,
+    )
+
+
+def _assert_cache_read_on_second_call(
+    client: PassthroughClient, key: str, model: str
+) -> None:
+    prefix = _cacheable_prefix()
+
+    first = unwrap(_cache_chat(client, key, model, prefix))
+    assert first.choices, f"{model}: first cache-priming call returned no choices: {first}"
+
+    read_tokens = 0
+    deadline = time.monotonic() + 30.0
+    while time.monotonic() < deadline:
+        second = unwrap(_cache_chat(client, key, model, prefix))
+        read_tokens = _cached_read_tokens(second.usage)
+        if read_tokens > 0:
+            break
+        time.sleep(3.0)
+
+    assert read_tokens > 0, (
+        f"{model}: second identical call reported no cache-read tokens "
+        f"({second.usage}); prompt caching did not take effect"
+    )
+
+
+class TestCacheControl:
+    @pytest.mark.covers(
+        "llm.chat_completions.bedrock_converse.prompt_cache_5m.nonstream.works",
+        exercised_on=[],
+    )
+    def test_bedrock_prompt_caching_reads_cache(
+        self, client: PassthroughClient, resources: ResourceManager
+    ) -> None:
+        model = f"e2e-bedrock-cache-{unique_marker()}"
+        model_id = client.gateway.create_model(
+            model,
+            LiteLLMParamsBody(model=BEDROCK_MODEL, aws_region_name="us-east-1"),
+        )
+        resources.defer(lambda: client.gateway.delete_model(model_id))
+        _assert_cache_read_on_second_call(client, resources.key(), model)
+
+    @pytest.mark.covers(
+        "llm.chat_completions.vertex.prompt_cache_5m.nonstream.works",
+        exercised_on=[],
+    )
+    def test_vertex_prompt_caching_reads_cache(
+        self, client: PassthroughClient, resources: ResourceManager
+    ) -> None:
+        model = f"e2e-vertex-cache-{unique_marker()}"
+        model_id = client.gateway.create_model(
+            model,
+            LiteLLMParamsBody(
+                model=VERTEX_MODEL,
+                vertex_project=os.environ.get("VERTEXAI_PROJECT"),
+                vertex_location="us-central1",
+                vertex_credentials=os.environ.get("VERTEXAI_CREDENTIALS"),
+            ),
+        )
+        resources.defer(lambda: client.gateway.delete_model(model_id))
+        _assert_cache_read_on_second_call(client, resources.key(), model)

tests/e2e/llm_translation/test_provider_features_e2e.py

@@ -1,4 +1,4 @@
-"""Live e2e for model-specific request features: service_tier and prompt caching.
+"""Live e2e for model-specific request features: service_tier.
 
 Each case asserts the feature took effect, not just a 200.
 
@@ -11,81 +11,22 @@ avoided here because it is capacity-constrained and returns a transient 429 when
 flex resources are unavailable. Bedrock and Vertex do not accept service_tier, so
 that cell is OpenAI-only by design.
 
-Prompt caching is asserted through provider prompt-cache usage tokens. The
-deterministic path is explicit ``cache_control`` on an Anthropic-family model
-(here Bedrock's Claude): a large cacheable prefix is sent twice and the second
-call must report ``cache_read_input_tokens > 0``. OpenAI and Gemini only offer
-implicit automatic caching, which does not deterministically produce a cache read
-within a test window (verified: repeated >3k-token prompts kept
-``prompt_tokens_details.cached_tokens`` at 0), so those caching cells are out of
-scope here and covered only by the explicit-cache-control Bedrock case.
+Prompt caching lives in test_cache_control.py.
 """
 
 from __future__ import annotations
 
 import pytest
-from pydantic import BaseModel, ConfigDict, Field
 
 from e2e_config import unique_marker
 from e2e_http import unwrap
 from lifecycle import ResourceManager
-from models import ChatBody, ChatMessage, ChatResponse, LiteLLMParamsBody
+from models import ChatBody, ChatMessage, LiteLLMParamsBody
 from passthrough_client import PassthroughClient
 
 pytestmark = pytest.mark.e2e
 
 SERVICE_TIER = "priority"
-CACHE_MIN_READ_TOKENS = 1
-
-
-class CacheControl(BaseModel):
-    type: str = "ephemeral"
-
-
-class CacheTextBlock(BaseModel):
-    type: str = "text"
-    text: str
-    cache_control: CacheControl | None = None
-
-
-class RichMessage(BaseModel):
-    role: str
-    content: list[CacheTextBlock]
-
-
-class CacheDirective(BaseModel):
-    """litellm per-request cache control. ``no-cache`` forces the proxy to skip its
-    own response cache and make a fresh provider call, so the second identical
-    request actually reaches Bedrock and reads the provider prompt cache instead of
-    being served the first response verbatim (which would report cache_read=0)."""
-
-    model_config = ConfigDict(populate_by_name=True)
-    no_cache: bool = Field(default=True, alias="no-cache")
-
-
-class CacheChatBody(BaseModel):
-    model: str
-    messages: list[RichMessage]
-    max_tokens: int
-    cache: CacheDirective = CacheDirective()
-
-
-def cacheable_prefix() -> str:
-    return (
-        "You are a policy compliance auditor. The following corpus is the immutable "
-        "reference the assistant must consult on every turn. "
-    ) + ("Clause: obey all safety, formatting, and citation rules exactly. " * 400)
-
-
-def post_chat(client: PassthroughClient, key: str, body: BaseModel) -> ChatResponse:
-    return unwrap(
-        client.gateway.transport.post(
-            "/chat/completions",
-            headers=client.gateway.transport.bearer(key),
-            json=body,
-            response_type=ChatResponse,
-        )
-    )
 
 
 class TestServiceTier:
@@ -120,50 +61,3 @@ class TestServiceTier:
             f"service_tier not honored: sent {SERVICE_TIER!r}, response reported "
             f"{response.service_tier!r} ({response})"
         )
-
-
-class TestPromptCaching:
-    @pytest.mark.covers(
-        "llm.chat_completions.bedrock_converse.prompt_cache_5m.nonstream.works",
-        exercised_on=[],
-    )
-    def test_bedrock_cache_control_produces_cache_read(
-        self, client: PassthroughClient, resources: ResourceManager
-    ) -> None:
-        model = f"e2e-bedrock-cache-{unique_marker()}"
-        model_id = client.gateway.create_model(
-            model,
-            LiteLLMParamsBody(
-                model="bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0",
-                aws_region_name="us-east-1",
-            ),
-        )
-        resources.defer(lambda: client.gateway.delete_model(model_id))
-        key = resources.key()
-
-        body = CacheChatBody(
-            model=model,
-            max_tokens=32,
-            messages=[
-                RichMessage(
-                    role="user",
-                    content=[
-                        CacheTextBlock(
-                            text=cacheable_prefix(), cache_control=CacheControl()
-                        ),
-                        CacheTextBlock(text="Answer in one word: acknowledged?"),
-                    ],
-                )
-            ],
-        )
-
-        first = post_chat(client, key, body)
-        assert first.usage is not None, f"first call reported no usage: {first}"
-
-        second = post_chat(client, key, body)
-        assert second.usage is not None, f"second call reported no usage: {second}"
-        cache_read = second.usage.cache_read_input_tokens
-        assert cache_read is not None and cache_read >= CACHE_MIN_READ_TOKENS, (
-            "second identical request did not read the prompt cache: "
-            f"cache_read_input_tokens={cache_read!r} (usage={second.usage})"
-        )

tests/e2e/management/management_client.py

@@ -6,10 +6,11 @@ llm-only key hitting a management route).
 
 from __future__ import annotations
 
+import time
 from dataclasses import dataclass
 
 from e2e_gateway import Gateway, build_gateway
-from e2e_http import NoBody, ProbeResult, StreamingResponse, unwrap
+from e2e_http import NoBody, ProbeResult, Result, StreamingResponse, Success, UnknownApiError, unwrap
 from models import (
     ChatBody,
     ChatMessage,
@@ -43,6 +44,8 @@ from models import (
 
 MODEL_ACCESS_DENIED_MARKER = "key_model_access_denied"
 ROUTE_NOT_ALLOWED_MARKER = "not allowed to call this route"
+_TEAM_READY_ATTEMPTS = 15
+_TEAM_READY_SLEEP_SECONDS = 0.4
 
 
 @dataclass(frozen=True, slots=True)
@@ -53,14 +56,26 @@ class ManagementClient:
         return self.gateway.generate_key(KeyGenerateBody(models=[], allowed_routes=["llm_api_routes"]))
 
     def update_key_models(self, key: str, models: list[str]) -> None:
-        _ = unwrap(
-            self.gateway.transport.post(
+        last: Result[NoBody] | None = None
+        for attempt in range(5):
+            last = self.gateway.transport.post(
                 "/key/update",
                 headers=self.gateway.transport.master,
                 json=KeyUpdateBody(key=key, models=models),
                 response_type=NoBody,
             )
-        )
+            match last:
+                case Success():
+                    return
+                case UnknownApiError(body=body) if (
+                    "connecting to redis" in body.lower() or "name resolution" in body.lower()
+                ):
+                    time.sleep(0.5 * (attempt + 1))
+                    continue
+                case _:
+                    break
+        assert last is not None
+        _ = unwrap(last)
 
     def delete_key_strict(self, key: str) -> None:
         """Strict delete for the act phase of a test: a failed delete is a hard
@@ -85,7 +100,7 @@ class ManagementClient:
         ).total_count
 
     def create_team(self, body: TeamNewBody) -> str:
-        return unwrap(
+        team_id = unwrap(
             self.gateway.transport.post(
                 "/team/new",
                 headers=self.gateway.transport.master,
@@ -93,6 +108,8 @@ class ManagementClient:
                 response_type=TeamNewResponse,
             )
         ).team_id
+        self._wait_for_team(team_id)
+        return team_id
 
     def delete_team(self, team_id: str) -> None:
         _ = self.gateway.transport.post(
@@ -115,15 +132,44 @@ class ManagementClient:
     def team_info_status(self, team_id: str) -> ProbeResult:
         return self.gateway.transport.probe("/team/info", params=TeamInfoParams(team_id=team_id))
 
+    def _wait_for_team(self, team_id: str) -> None:
+        last: Result[TeamInfoResponse] | None = None
+        for _ in range(_TEAM_READY_ATTEMPTS):
+            last = self.gateway.transport.get(
+                "/team/info",
+                headers=self.gateway.transport.master,
+                params=TeamInfoParams(team_id=team_id),
+                response_type=TeamInfoResponse,
+            )
+            match last:
+                case Success():
+                    return
+                case _:
+                    time.sleep(_TEAM_READY_SLEEP_SECONDS)
+        assert last is not None
+        _ = unwrap(last)
+
     def add_team_member(self, team_id: str, user_id: str) -> None:
-        _ = unwrap(
-            self.gateway.transport.post(
+        last: Result[NoBody] | None = None
+        for attempt in range(_TEAM_READY_ATTEMPTS):
+            last = self.gateway.transport.post(
                 "/team/member_add",
                 headers=self.gateway.transport.master,
                 json=TeamMemberAddBody(team_id=team_id, member=TeamMemberEntry(role="user", user_id=user_id)),
                 response_type=NoBody,
             )
-        )
+            match last:
+                case Success():
+                    return
+                case UnknownApiError(body=body) if (
+                    "doesn't exist" in body and attempt + 1 < _TEAM_READY_ATTEMPTS
+                ):
+                    time.sleep(_TEAM_READY_SLEEP_SECONDS)
+                    continue
+                case _:
+                    break
+        assert last is not None
+        _ = unwrap(last)
 
     def delete_team_member(self, team_id: str, user_id: str) -> None:
         _ = unwrap(

tests/e2e/models.py

@@ -377,10 +377,13 @@ class LiteLLMParamsBody(BaseModel):
     api_base: str | None = None
     api_version: str | None = None
     realtime_protocol: str | None = None
+    aws_access_key_id: str | None = None
+    aws_secret_access_key: str | None = None
     aws_region_name: str | None = None
     vertex_project: str | None = None
     vertex_location: str | None = None
     vertex_credentials: str | None = None
+    gcs_bucket_name: str | None = None
     bucket_name: str | None = None
     s3_bucket_name: str | None = None
     s3_region_name: str | None = None

tests/proxy_behavior/management/test_team_update.py

@@ -105,27 +105,35 @@ async def test_team_update_authz_matrix(
         assert row.team_alias != MARKER_ALIAS, "denied but team mutated"
 
 
-async def test_team_update_requires_proxy_admin_without_org_context(
+async def test_team_update_org_admin_resolved_from_team_without_org_context(
     proxy_client, prisma, scratch, world
 ):
-    """With no organization_id in the body the route gate has no org context
-    and falls back to proxy-admin-only: an org admin of the team's own org
-    is 401, PROXY_ADMIN is 200."""
+    """With no organization_id in the body the route gate resolves the target
+    team's org from team_id, so an org admin of the team's own org is allowed
+    (200), same as PROXY_ADMIN. A team admin of that same team stays denied
+    (401): the resolution grants org admins access, not team admins."""
     await _seed_target(prisma, world, "alpha", scratch.prefix)
 
-    denied = await proxy_client.post(
+    allowed_org_admin = await proxy_client.post(
         "/team/update",
         headers={"Authorization": f"Bearer {world.keys[Actor.ORG_ADMIN].cleartext}"},
         json={"team_id": scratch.prefix, "team_alias": MARKER_ALIAS},
     )
-    assert denied.status_code == 401, denied.text
+    assert allowed_org_admin.status_code == 200, allowed_org_admin.text
 
-    allowed = await proxy_client.post(
+    allowed_proxy_admin = await proxy_client.post(
         "/team/update",
         headers={"Authorization": f"Bearer {world.keys[Actor.PROXY_ADMIN].cleartext}"},
         json={"team_id": scratch.prefix, "team_alias": MARKER_ALIAS},
     )
-    assert allowed.status_code == 200, allowed.text
+    assert allowed_proxy_admin.status_code == 200, allowed_proxy_admin.text
+
+    denied_team_admin = await proxy_client.post(
+        "/team/update",
+        headers={"Authorization": f"Bearer {world.keys[Actor.TEAM_ADMIN].cleartext}"},
+        json={"team_id": scratch.prefix, "team_alias": MARKER_ALIAS},
+    )
+    assert denied_team_admin.status_code == 401, denied_team_admin.text
 
 
 # Relocation gate — moving a team to a different org. The scratch team starts

tests/test_litellm/integrations/test_custom_guardrail.py

@@ -833,6 +833,173 @@ class TestGuardrailSensitiveFieldStripping:
         assert "sk-secret" not in serialized
 
 
+class TestGuardrailResponseCredentialMasking:
+    """LIT-4314 issue B regression: credentials embedded in guardrail_response
+    (via team callback_vars flowing through data["metadata"]) must be masked at
+    the construction seam so every downstream sink (SpendLogs, OTel, Langfuse,
+    custom loggers) sees masked values rather than plaintext.
+    """
+
+    def _make_guardrail(self):
+        from litellm.types.guardrails import GuardrailEventHooks
+
+        return CustomGuardrail(
+            guardrail_name="test_guardrail",
+            event_hook=GuardrailEventHooks.pre_call,
+        )
+
+    def test_callback_vars_api_key_is_masked(self):
+        import json
+
+        guardrail = self._make_guardrail()
+        request_data: dict = {"metadata": {}}
+        plaintext_key = "lsv2_pt_abcdef1234567890"
+
+        guardrail.add_standard_logging_guardrail_information_to_request_data(
+            guardrail_json_response={
+                "model": "gpt-4o-mini",
+                "messages": [{"role": "user", "content": "hi"}],
+                "metadata_snapshot": {
+                    "callback_vars": {
+                        "langsmith_api_key": plaintext_key,
+                        "langsmith_project": "proj-name",
+                    }
+                },
+            },
+            request_data=request_data,
+            guardrail_status="success",
+            duration=1.0,
+        )
+
+        logged = request_data["metadata"]["standard_logging_guardrail_information"][0][
+            "guardrail_response"
+        ]
+
+        masked_key = logged["metadata_snapshot"]["callback_vars"]["langsmith_api_key"]
+        assert masked_key != plaintext_key
+        assert "*" in masked_key
+        assert plaintext_key not in json.dumps(request_data)
+
+        assert logged["model"] == "gpt-4o-mini"
+        assert logged["messages"] == [{"role": "user", "content": "hi"}]
+        assert (
+            logged["metadata_snapshot"]["callback_vars"]["langsmith_project"]
+            == "proj-name"
+        )
+
+    def test_nested_user_api_key_auth_metadata_is_masked(self):
+        import json
+
+        guardrail = self._make_guardrail()
+        request_data: dict = {"metadata": {}}
+        token_value = "1b01552f6e52e0d41963dd6a185bd6b074624e330999534ca7ff5adfdf622dfc"
+
+        guardrail.add_standard_logging_guardrail_information_to_request_data(
+            guardrail_json_response={
+                "evaluated_metadata": {
+                    "user_api_key_auth": {
+                        "token": token_value,
+                        "api_key": token_value,
+                        "metadata": {
+                            "callback_vars": {
+                                "langsmith_api_key": "lsv2_pt_super_secret_value_1234",
+                            }
+                        },
+                    }
+                }
+            },
+            request_data=request_data,
+            guardrail_status="success",
+        )
+
+        serialized = json.dumps(request_data)
+        assert token_value not in serialized
+        assert "lsv2_pt_super_secret_value_1234" not in serialized
+
+    def test_secret_fields_pop_still_runs(self):
+        import json
+
+        guardrail = self._make_guardrail()
+        request_data: dict = {"metadata": {}}
+
+        guardrail.add_standard_logging_guardrail_information_to_request_data(
+            guardrail_json_response={
+                "model": "gpt-4",
+                "secret_fields": {
+                    "raw_headers": {
+                        "authorization": "Bearer sk-live-should-not-appear",
+                    }
+                },
+            },
+            request_data=request_data,
+            guardrail_status="success",
+        )
+
+        serialized = json.dumps(request_data)
+        assert "secret_fields" not in serialized
+        assert "sk-live-should-not-appear" not in serialized
+
+    def test_match_and_regex_redaction_still_runs(self):
+        guardrail = self._make_guardrail()
+        request_data: dict = {"metadata": {}}
+
+        guardrail.add_standard_logging_guardrail_information_to_request_data(
+            guardrail_json_response={
+                "filters": [{"regex": r"\d{3}-\d{2}-\d{4}", "action": "BLOCKED"}]
+            },
+            request_data=request_data,
+            guardrail_status="success",
+        )
+
+        slg = request_data["metadata"]["standard_logging_guardrail_information"][0]
+        assert slg["guardrail_response"]["filters"][0]["regex"] == "[REDACTED]"
+
+    def test_scalar_types_pass_through_unchanged(self):
+        guardrail = self._make_guardrail()
+        request_data: dict = {"metadata": {}}
+
+        guardrail.add_standard_logging_guardrail_information_to_request_data(
+            guardrail_json_response={
+                "flagged": True,
+                "score": 0.94,
+                "tokens_used": 42,
+                "categories": ["pii", "toxicity"],
+            },
+            request_data=request_data,
+            guardrail_status="success",
+        )
+
+        logged = request_data["metadata"]["standard_logging_guardrail_information"][0][
+            "guardrail_response"
+        ]
+        assert logged["flagged"] is True
+        assert logged["score"] == 0.94
+        assert logged["tokens_used"] == 42
+        assert logged["categories"] == ["pii", "toxicity"]
+
+    def test_masking_reveals_prefix_and_suffix(self):
+        guardrail = self._make_guardrail()
+        request_data: dict = {"metadata": {}}
+        plaintext = "lsv2_pt_abcdef1234567890"
+
+        guardrail.add_standard_logging_guardrail_information_to_request_data(
+            guardrail_json_response={
+                "metadata_snapshot": {
+                    "callback_vars": {"langsmith_api_key": plaintext}
+                }
+            },
+            request_data=request_data,
+            guardrail_status="success",
+        )
+
+        masked = request_data["metadata"]["standard_logging_guardrail_information"][0][
+            "guardrail_response"
+        ]["metadata_snapshot"]["callback_vars"]["langsmith_api_key"]
+        assert masked != plaintext
+        assert masked.startswith(plaintext[:4])
+        assert masked.endswith(plaintext[-4:])
+
+
 class TestCustomGuardrailPassthroughSupport:
     """Tests for passthrough endpoint guardrail support - Issue fixes."""
 

tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py

@@ -562,6 +562,49 @@ def test_generic_cost_per_token_gpt56(
     assert round(completion_cost, 10) == round(output_cost * completion_tokens, 10)
 
 
+@pytest.mark.parametrize(
+    "model,input_cost,output_cost,cache_read_cost",
+    [
+        ("azure/gpt-5.6", 5e-6, 3e-5, 5e-7),
+        ("azure/gpt-5.6-sol", 5e-6, 3e-5, 5e-7),
+        ("azure/gpt-5.6-terra", 2.5e-6, 1.5e-5, 2.5e-7),
+        ("azure/gpt-5.6-luna", 1e-6, 6e-6, 1e-7),
+        ("azure/us/gpt-5.6", 5.5e-6, 3.3e-5, 5.5e-7),
+        ("azure/eu/gpt-5.6-terra", 2.75e-6, 1.65e-5, 2.75e-7),
+        ("azure/eu/gpt-5.6-luna", 1.1e-6, 6.6e-6, 1.1e-7),
+    ],
+)
+def test_generic_cost_per_token_azure_gpt56(
+    model, input_cost, output_cost, cache_read_cost
+):
+    """Azure gpt-5.6 (global + us/eu regional): pricing mirrors the openai
+    family for global deployments and carries the standard 10% regional uplift.
+    """
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
+
+    model_cost_map = litellm.model_cost[model]
+    assert model_cost_map["litellm_provider"] == "azure"
+    assert model_cost_map["input_cost_per_token"] == input_cost
+    assert model_cost_map["output_cost_per_token"] == output_cost
+    assert model_cost_map["cache_read_input_token_cost"] == cache_read_cost
+
+    prompt_tokens = 1000
+    completion_tokens = 500
+    usage = Usage(
+        prompt_tokens=prompt_tokens,
+        completion_tokens=completion_tokens,
+        total_tokens=prompt_tokens + completion_tokens,
+    )
+    prompt_cost, completion_cost = generic_cost_per_token(
+        model=model,
+        usage=usage,
+        custom_llm_provider="azure",
+    )
+    assert round(prompt_cost, 10) == round(input_cost * prompt_tokens, 10)
+    assert round(completion_cost, 10) == round(output_cost * completion_tokens, 10)
+
+
 @pytest.mark.parametrize(
     "model,expected_none,expected_xhigh,expected_minimal",
     [

tests/test_litellm/litellm_core_utils/test_sensitive_data_masker.py

@@ -240,3 +240,78 @@ def test_mask_sensitive_structure_masks_credentials_nested_in_config_shape():
         [{"primary-group": [{"model": "gpt-4o", "api_key": secret}]}]
     )
     assert secret not in str(masked)
+
+
+def test_mask_credentials_in_payload_preserves_none_and_scalars():
+    """The payload variant does not distort JSON-shaped values: None stays None,
+    ints/floats/bools stay themselves, lists stay lists. This is what makes it
+    safe for logging pipelines that persist the record verbatim."""
+    from litellm.litellm_core_utils.sensitive_data_masker import mask_credentials_in_payload
+
+    result = mask_credentials_in_payload(
+        {
+            "reason": None,
+            "confidence": 0.42,
+            "flagged": True,
+            "tokens_used": 17,
+            "categories": ["pii", "toxicity"],
+            "nested": {"end_user_id": None},
+        }
+    )
+    assert result == {
+        "reason": None,
+        "confidence": 0.42,
+        "flagged": True,
+        "tokens_used": 17,
+        "categories": ["pii", "toxicity"],
+        "nested": {"end_user_id": None},
+    }
+
+
+def test_mask_credentials_in_payload_masks_inside_pydantic_models():
+    """A Pydantic model reached during the walk gets dumped to a dict so its
+    sensitive-named string fields are masked. Without this the credentials
+    inside a nested ``UserAPIKeyAuth`` in a guardrail_response reach the
+    logging pipeline unmasked once JSON serialization flattens it."""
+    from pydantic import BaseModel
+
+    from litellm.litellm_core_utils.sensitive_data_masker import mask_credentials_in_payload
+
+    class Auth(BaseModel):
+        token: str = "1b01552f6e52e0d41963dd6a185bd6b074624e330999534ca7ff5adfdf622dfc"
+        team_alias: str = "acme"
+
+    result = mask_credentials_in_payload({"user_api_key_auth": Auth()})
+    auth_dict = result["user_api_key_auth"]
+    assert isinstance(auth_dict, dict)
+    assert auth_dict["team_alias"] == "acme"
+    assert (
+        auth_dict["token"]
+        != "1b01552f6e52e0d41963dd6a185bd6b074624e330999534ca7ff5adfdf622dfc"
+    )
+    assert "*" in auth_dict["token"]
+
+
+def test_mask_credentials_in_payload_masks_only_sensitive_string_leaves():
+    """Sensitive-named string leaves get masked; sibling non-string values
+    (including None) under the same key stay verbatim."""
+    from litellm.litellm_core_utils.sensitive_data_masker import mask_credentials_in_payload
+
+    plaintext = "lsv2_pt_abcdef1234567890"
+    result = mask_credentials_in_payload(
+        {
+            "model": "gpt-4o-mini",
+            "callback_vars": {
+                "langsmith_api_key": plaintext,
+                "langsmith_project": "proj",
+                "extra_token_count": 5,
+            },
+        }
+    )
+    assert result["model"] == "gpt-4o-mini"
+    assert result["callback_vars"]["langsmith_project"] == "proj"
+    assert result["callback_vars"]["extra_token_count"] == 5
+    masked = result["callback_vars"]["langsmith_api_key"]
+    assert masked != plaintext
+    assert masked.startswith(plaintext[:4])
+    assert masked.endswith(plaintext[-4:])

tests/test_litellm/llms/openai_like/test_meta_provider.py

@@ -0,0 +1,224 @@
+"""
+Tests for the Meta Model API (Muse Spark) provider configuration and integration.
+"""
+
+import litellm
+
+
+class TestMetaProviderConfig:
+    def test_meta_in_provider_list(self):
+        from litellm import LlmProviders
+
+        assert hasattr(LlmProviders, "META")
+        assert LlmProviders.META.value == "meta"
+        assert "meta" in litellm.provider_list
+
+    def test_meta_json_config_exists(self):
+        from litellm.llms.openai_like.json_loader import JSONProviderRegistry
+
+        assert JSONProviderRegistry.exists("meta")
+
+        meta = JSONProviderRegistry.get("meta")
+        assert meta is not None
+        assert meta.base_url == "https://api.meta.ai/v1"
+        assert meta.api_key_env == "META_API_KEY"
+        assert meta.api_base_env == "META_API_BASE"
+
+    def test_meta_supports_responses_api(self):
+        from litellm.llms.openai_like.json_loader import JSONProviderRegistry
+
+        assert JSONProviderRegistry.supports_responses_api("meta")
+
+    def test_meta_in_openai_compatible_providers(self):
+        from litellm.constants import openai_compatible_providers
+
+        assert "meta" in openai_compatible_providers
+
+    def test_meta_provider_resolution(self):
+        from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+        model, provider, api_key, api_base = get_llm_provider(
+            model="meta/muse-spark-1.1",
+            custom_llm_provider=None,
+            api_base=None,
+            api_key="sk-test",
+        )
+
+        assert model == "muse-spark-1.1"
+        assert provider == "meta"
+        assert api_base == "https://api.meta.ai/v1"
+
+    def test_meta_api_base_override(self):
+        from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+        model, provider, api_key, api_base = get_llm_provider(
+            model="meta/muse-spark-1.1",
+            custom_llm_provider=None,
+            api_base="https://custom.meta.ai/v1",
+            api_key="sk-test",
+        )
+
+        assert provider == "meta"
+        assert api_base == "https://custom.meta.ai/v1"
+        assert api_key == "sk-test"
+
+    def test_meta_url_autodetection(self):
+        from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+        model, provider, api_key, api_base = get_llm_provider(
+            model="muse-spark-1.1",
+            custom_llm_provider=None,
+            api_base="https://api.meta.ai/v1",
+            api_key=None,
+        )
+        assert provider == "meta"
+        assert api_base == "https://api.meta.ai/v1"
+
+    def test_meta_router_config(self):
+        from litellm import Router
+
+        router = Router(
+            model_list=[
+                {
+                    "model_name": "muse-spark",
+                    "litellm_params": {
+                        "model": "meta/muse-spark-1.1",
+                        "api_key": "test-key",
+                    },
+                }
+            ]
+        )
+
+        assert len(router.model_list) == 1
+        assert router.model_list[0]["model_name"] == "muse-spark"
+
+
+class TestMetaReasoningParams:
+    def test_muse_spark_supports_reasoning_effort(self):
+        params = litellm.get_supported_openai_params(
+            model="muse-spark-1.1", custom_llm_provider="meta"
+        )
+        assert params is not None
+        assert "reasoning_effort" in params
+
+    def test_reasoning_effort_mapped_through(self):
+        cfg = litellm.ProviderConfigManager.get_provider_chat_config(
+            model="muse-spark-1.1", provider=litellm.LlmProviders.META
+        )
+        assert cfg is not None
+        mapped = cfg.map_openai_params(
+            non_default_params={"reasoning_effort": "xhigh"},
+            optional_params={},
+            model="muse-spark-1.1",
+            drop_params=False,
+        )
+        assert mapped["reasoning_effort"] == "xhigh"
+
+    def test_reasoning_effort_gated_on_capability(self):
+        """A meta model without reasoning metadata must not advertise reasoning_effort."""
+        params = litellm.get_supported_openai_params(
+            model="some-non-reasoning-model", custom_llm_provider="meta"
+        )
+        assert params is not None
+        assert "reasoning_effort" not in params
+
+
+class TestMetaAnthropicMessages:
+    def test_meta_resolves_native_messages_config(self):
+        from litellm.llms.openai_like.messages.transformation import (
+            JSONProviderAnthropicMessagesConfig,
+        )
+
+        cfg = litellm.ProviderConfigManager.get_provider_anthropic_messages_config(
+            model="muse-spark-1.1", provider=litellm.LlmProviders.META
+        )
+        assert isinstance(cfg, JSONProviderAnthropicMessagesConfig)
+
+    def test_json_provider_without_messages_endpoint_resolves_none(self):
+        cfg = litellm.ProviderConfigManager.get_provider_anthropic_messages_config(
+            model="some-model", provider=litellm.LlmProviders.PINSTRIPES
+        )
+        assert cfg is None
+
+    def test_complete_url_defaults_to_meta_base(self):
+        from litellm.llms.openai_like.json_loader import JSONProviderRegistry
+        from litellm.llms.openai_like.messages.transformation import (
+            JSONProviderAnthropicMessagesConfig,
+        )
+
+        provider = JSONProviderRegistry.get("meta")
+        assert provider is not None
+        cfg = JSONProviderAnthropicMessagesConfig(provider)
+
+        url = cfg.get_complete_url(
+            api_base=None,
+            api_key="sk-test",
+            model="muse-spark-1.1",
+            optional_params={},
+            litellm_params={},
+        )
+        assert url == "https://api.meta.ai/v1/messages"
+
+        override_url = cfg.get_complete_url(
+            api_base="https://custom.meta.ai/v1",
+            api_key="sk-test",
+            model="muse-spark-1.1",
+            optional_params={},
+            litellm_params={},
+        )
+        assert override_url == "https://custom.meta.ai/v1/messages"
+
+    def test_api_key_resolved_from_env(self, monkeypatch):
+        from litellm.llms.openai_like.json_loader import JSONProviderRegistry
+        from litellm.llms.openai_like.messages.transformation import (
+            JSONProviderAnthropicMessagesConfig,
+        )
+
+        monkeypatch.setenv("META_API_KEY", "sk-env-key")
+        provider = JSONProviderRegistry.get("meta")
+        assert provider is not None
+        cfg = JSONProviderAnthropicMessagesConfig(provider)
+
+        headers, _ = cfg.validate_anthropic_messages_environment(
+            headers={},
+            model="muse-spark-1.1",
+            messages=[{"role": "user", "content": "hi"}],
+            optional_params={},
+            litellm_params={},
+            api_key=None,
+            api_base=None,
+        )
+        assert headers["authorization"] == "Bearer sk-env-key"
+        assert headers["anthropic-version"] == "2023-06-01"
+
+
+class TestMuseSparkModelInfo:
+    def test_muse_spark_pricing_and_capabilities(self):
+        info = litellm.get_model_info("meta/muse-spark-1.1")
+
+        assert info["litellm_provider"] == "meta"
+        assert info["input_cost_per_token"] == 1.25e-06
+        assert info["output_cost_per_token"] == 4.25e-06
+        assert info["cache_read_input_token_cost"] == 1.5e-07
+        assert info["max_input_tokens"] == 1048576
+        assert info["supports_reasoning"] is True
+        assert info["supports_web_search"] is True
+        assert info["supports_vision"] is True
+        assert info["supports_function_calling"] is True
+        assert info["supports_prompt_caching"] is True
+
+    def test_muse_spark_cost_calculation(self):
+        from litellm import completion_cost
+        from litellm.types.utils import ModelResponse, Usage
+
+        response = ModelResponse(
+            model="muse-spark-1.1",
+            usage=Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500),
+        )
+        cost = completion_cost(
+            completion_response=response,
+            model="meta/muse-spark-1.1",
+            custom_llm_provider="meta",
+        )
+        expected = 1000 * 1.25e-06 + 500 * 4.25e-06
+        assert abs(cost - expected) < 1e-12

tests/test_litellm/proxy/_experimental/mcp_server/outbound_credentials/test_resolver.py

@@ -149,6 +149,29 @@ async def test_authorization_code_isolates_by_subject():
     assert isinstance(bob, Error) and bob.error.tag == "unauthorized"
 
 
+@pytest.mark.asyncio
+async def test_authorization_code_isolates_by_server_id_even_when_servers_share_a_url():
+    """A token stored for one server must be invisible to a different server_id pointing at the
+    same upstream URL: credentials bind to the server entry they were authorized for, so a
+    recreated or duplicated server starts unauthorized instead of inheriting the old grant. Guards
+    against any future token lookup keyed on the resource URL instead of (user_id, server_id) --
+    both the egress resolve and the has_user_token discovery check must agree."""
+    shared_url = "https://upstream.example.com"
+    store = _FakeTokenStore({("alice", "server-a"): OAuthToken(access_token="at-alice")})
+    provider = UpstreamCredentialProvider(oauth_token_store=store)
+    subject = Subject(tenant_id="", subject_id="alice")
+    spec_a = ServerSpec(server_id="server-a", resource=shared_url, config=AuthorizationCodeConfig())
+    spec_b = ServerSpec(server_id="server-b", resource=shared_url, config=AuthorizationCodeConfig())
+
+    granted = await provider.resolve_credentials(subject, spec_a)
+    fresh = await provider.resolve_credentials(subject, spec_b)
+
+    assert isinstance(granted, Ok) and _emitted(granted.ok)["Authorization"] == "Bearer at-alice"
+    assert isinstance(fresh, Error) and fresh.error.tag == "unauthorized"
+    assert await provider.has_user_token(subject, spec_a) is True
+    assert await provider.has_user_token(subject, spec_b) is False
+
+
 @pytest.mark.asyncio
 async def test_has_user_token_reflects_the_stored_token():
     present = UpstreamCredentialProvider(

tests/test_litellm/proxy/_experimental/mcp_server/test_db_credentials.py

@@ -11,6 +11,7 @@ keeps a plain-base64 fallback on read so existing rows continue to work.
 import base64
 import json
 from datetime import datetime, timedelta, timezone
+from types import SimpleNamespace
 from unittest.mock import AsyncMock, MagicMock
 
 import pytest
@@ -63,6 +64,264 @@ def _legacy_row(payload: str):
     return row
 
 
+def _identity_server(**overrides):
+    base = dict(
+        url="https://up.example.com/mcp",
+        auth_type="oauth2",
+        oauth2_flow="authorization_code",
+        authorization_url="https://idp.example.com/authorize",
+        token_url="https://idp.example.com/token",
+        registration_url="https://idp.example.com/register",
+        credentials={"client_id": "cid", "client_secret": "csec", "scopes": ["a"]},
+        server_name="srv",
+        description="d",
+    )
+    base.update(overrides)
+    return SimpleNamespace(**base)
+
+
+@pytest.mark.parametrize(
+    "overrides",
+    [
+        {"url": "https://other.example.com/mcp"},
+        {"spec_path": "https://up.example.com/openapi.json"},
+        {"auth_type": "oauth_delegate"},
+        {"oauth2_flow": "client_credentials"},
+        {"authorization_url": "https://other.example.com/authorize"},
+        {"token_url": "https://other.example.com/token"},
+        {"registration_url": "https://other.example.com/register"},
+        {"credentials": {"client_id": "new", "client_secret": "csec", "scopes": ["a"]}},
+        {"credentials": {"client_id": "cid", "client_secret": "rotated", "scopes": ["a"]}},
+        {"credentials": {"client_id": "cid", "client_secret": "csec", "scopes": ["b"]}},
+    ],
+)
+def test_mcp_oauth_token_identity_changes_on_mint_relevant_fields(overrides):
+    from litellm.proxy._experimental.mcp_server.db import mcp_oauth_token_identity
+
+    assert mcp_oauth_token_identity(_identity_server()) != mcp_oauth_token_identity(_identity_server(**overrides))
+
+
+@pytest.mark.parametrize(
+    "overrides",
+    [
+        {"server_name": "renamed"},
+        {"description": "changed"},
+    ],
+)
+def test_mcp_oauth_token_identity_stable_on_non_mint_fields(overrides):
+    from litellm.proxy._experimental.mcp_server.db import mcp_oauth_token_identity
+
+    assert mcp_oauth_token_identity(_identity_server()) == mcp_oauth_token_identity(_identity_server(**overrides))
+
+
+def _encrypted_creds_json(client_id: str = "cid", client_secret: str = "csec") -> str:
+    from litellm.proxy._experimental.mcp_server.db import encrypt_credentials
+
+    encrypted = encrypt_credentials(
+        credentials={"client_id": client_id, "client_secret": client_secret, "scopes": ["a"]},
+        encryption_key=None,
+    )
+    return json.dumps(encrypted)
+
+
+def test_mcp_oauth_token_identity_stable_across_reencryption():
+    """Stored client_id/client_secret are NaCl-encrypted with a fresh nonce on every write, so two
+    saves of the SAME plaintext produce different ciphertext. The identity must compare decrypted
+    values; comparing ciphertext would flag every routine save as a mint-relevant change and purge
+    per-user tokens that are still valid."""
+    from litellm.proxy._experimental.mcp_server.db import mcp_oauth_token_identity
+
+    first = _encrypted_creds_json()
+    second = _encrypted_creds_json()
+    assert first != second
+
+    assert mcp_oauth_token_identity(_identity_server(credentials=first)) == mcp_oauth_token_identity(
+        _identity_server(credentials=second)
+    )
+
+
+def test_mcp_oauth_token_identity_detects_change_under_encryption():
+    from litellm.proxy._experimental.mcp_server.db import mcp_oauth_token_identity
+
+    unchanged = _identity_server(credentials=_encrypted_creds_json())
+    changed = _identity_server(credentials=_encrypted_creds_json(client_id="other"))
+    assert mcp_oauth_token_identity(unchanged) != mcp_oauth_token_identity(changed)
+
+
+def _oauth_row(user_id: str, server_id: str = "srv-1"):
+    """A stored per-user OAuth token row (payload tagged type=oauth2, legacy plain-base64 encoding)."""
+    row = _legacy_row(json.dumps({"type": "oauth2", "access_token": "tok-" + user_id}))
+    row.user_id = user_id
+    row.server_id = server_id
+    return row
+
+
+def _byok_row(user_id: str, server_id: str = "srv-1"):
+    """A stored BYOK API key row: the same column, but the payload is a plain string, not OAuth JSON."""
+    row = _legacy_row("sk-byok-" + user_id)
+    row.user_id = user_id
+    row.server_id = server_id
+    return row
+
+
+@pytest.mark.asyncio
+async def test_purge_user_oauth_credentials_for_server_invalidates_each_user():
+    """The purge must route each (user, server) row through the invalidator exactly once."""
+    from litellm.proxy._experimental.mcp_server.db import purge_user_oauth_credentials_for_server
+
+    prisma = MagicMock()
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[_oauth_row("alice"), _oauth_row("bob")])
+    prisma.db.litellm_mcpusercredentials.delete_many = AsyncMock(return_value=2)
+
+    invalidations = []
+
+    async def record_invalidation(user_id: str, server_id: str) -> None:
+        invalidations.append((user_id, server_id))
+
+    purged = await purge_user_oauth_credentials_for_server(prisma, "srv-1", invalidate_token_cache=record_invalidation)
+
+    assert purged == 2
+    prisma.db.litellm_mcpusercredentials.delete_many.assert_awaited_once_with(
+        where={"server_id": "srv-1", "user_id": {"in": ["alice", "bob"]}}
+    )
+    assert set(invalidations) == {("alice", "srv-1"), ("bob", "srv-1")}
+
+
+@pytest.mark.asyncio
+async def test_purge_user_oauth_credentials_for_server_spares_byok_rows():
+    """Regression: the purge used to delete_many on server_id alone, wiping BYOK API keys that share
+    the LiteLLM_MCPUserCredentials table. Only rows holding an OAuth2 payload may be deleted (one
+    batched query filtered to their user_ids), and only their users' token caches invalidated."""
+    from litellm.proxy._experimental.mcp_server.db import purge_user_oauth_credentials_for_server
+
+    prisma = MagicMock()
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[_byok_row("carol"), _oauth_row("alice")])
+    prisma.db.litellm_mcpusercredentials.delete_many = AsyncMock(return_value=1)
+
+    invalidations = []
+
+    async def record_invalidation(user_id: str, server_id: str) -> None:
+        invalidations.append((user_id, server_id))
+
+    purged = await purge_user_oauth_credentials_for_server(prisma, "srv-1", invalidate_token_cache=record_invalidation)
+
+    assert purged == 1
+    prisma.db.litellm_mcpusercredentials.delete_many.assert_awaited_once_with(
+        where={"server_id": "srv-1", "user_id": {"in": ["alice"]}}
+    )
+    assert invalidations == [("alice", "srv-1")]
+
+
+@pytest.mark.asyncio
+async def test_purge_user_oauth_credentials_for_server_all_byok_is_noop():
+    """An api_key (BYOK-only) server whose identity tuple changes (e.g. its url) must purge nothing."""
+    from litellm.proxy._experimental.mcp_server.db import purge_user_oauth_credentials_for_server
+
+    prisma = MagicMock()
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[_byok_row("carol"), _byok_row("dave")])
+    prisma.db.litellm_mcpusercredentials.delete_many = AsyncMock()
+
+    purged = await purge_user_oauth_credentials_for_server(prisma, "srv-1")
+
+    assert purged == 0
+    prisma.db.litellm_mcpusercredentials.delete_many.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_purge_user_oauth_credentials_for_server_defaults_to_manager_invalidator(monkeypatch):
+    """When no invalidator is injected, the purge must resolve to the manager's shared
+    invalidate_user_oauth_token_cache, the single point covering both the legacy per-user token cache
+    and the v2 per-user OAuth token store; a wrong or no-op default silently leaves every cache
+    serving tokens minted for the superseded config."""
+    from litellm.proxy._experimental.mcp_server import mcp_server_manager
+    from litellm.proxy._experimental.mcp_server.db import purge_user_oauth_credentials_for_server
+
+    prisma = MagicMock()
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[_oauth_row("alice")])
+    prisma.db.litellm_mcpusercredentials.delete_many = AsyncMock(return_value=1)
+
+    shared_invalidator = AsyncMock()
+    monkeypatch.setattr(
+        mcp_server_manager.global_mcp_server_manager,
+        "invalidate_user_oauth_token_cache",
+        shared_invalidator,
+    )
+
+    purged = await purge_user_oauth_credentials_for_server(prisma, "srv-1")
+
+    assert purged == 1
+    shared_invalidator.assert_awaited_once_with("alice", "srv-1")
+
+
+@pytest.mark.asyncio
+async def test_purge_user_oauth_credentials_for_server_logs_raced_rows(monkeypatch):
+    from litellm.proxy._experimental.mcp_server import db as db_module
+    from litellm.proxy._experimental.mcp_server.db import purge_user_oauth_credentials_for_server
+
+    prisma = MagicMock()
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[_oauth_row("alice")])
+    prisma.db.litellm_mcpusercredentials.delete_many = AsyncMock(return_value=0)
+    warning = MagicMock()
+    monkeypatch.setattr(db_module.verbose_proxy_logger, "warning", warning)
+
+    purged = await purge_user_oauth_credentials_for_server(prisma, "srv-1", invalidate_token_cache=AsyncMock())
+
+    assert purged == 0
+    warning.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_delete_mcp_server_invalidates_cached_tokens_for_enumerated_users():
+    """Deleting a server must invalidate each enumerated user's cached per-user token: the caches are
+    keyed by (user_id, server_id), so a re-created server reusing the same server_id would otherwise
+    serve tokens minted for the deleted server until TTL."""
+    from litellm.proxy._experimental.mcp_server.db import delete_mcp_server
+
+    prisma = MagicMock()
+    prisma.db.litellm_mcpservertable.delete = AsyncMock(return_value=MagicMock(server_id="srv-1"))
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[_oauth_row("alice"), _byok_row("bob")])
+    prisma.db.litellm_mcpusercredentials.delete_many = AsyncMock(return_value=2)
+    prisma.db.litellm_mcpuserenvvars.delete_many = AsyncMock(return_value=0)
+
+    invalidations = []
+
+    async def record_invalidation(user_id: str, server_id: str) -> None:
+        invalidations.append((user_id, server_id))
+
+    deleted = await delete_mcp_server(prisma, "srv-1", invalidate_token_cache=record_invalidation)
+
+    assert deleted is not None
+    assert set(invalidations) == {("alice", "srv-1"), ("bob", "srv-1")}
+
+
+@pytest.mark.asyncio
+async def test_delete_mcp_server_returns_none_without_cleanup_when_server_missing():
+    from litellm.proxy._experimental.mcp_server.db import delete_mcp_server
+
+    prisma = MagicMock()
+    prisma.db.litellm_mcpservertable.delete = AsyncMock(return_value=None)
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock()
+
+    deleted = await delete_mcp_server(prisma, "srv-1", invalidate_token_cache=AsyncMock())
+
+    assert deleted is None
+    prisma.db.litellm_mcpusercredentials.find_many.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_purge_user_oauth_credentials_for_server_noop_when_empty():
+    from litellm.proxy._experimental.mcp_server.db import purge_user_oauth_credentials_for_server
+
+    prisma = MagicMock()
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[])
+    prisma.db.litellm_mcpusercredentials.delete_many = AsyncMock()
+
+    purged = await purge_user_oauth_credentials_for_server(prisma, "srv-1")
+
+    assert purged == 0
+    prisma.db.litellm_mcpusercredentials.delete_many.assert_not_awaited()
+
+
 def _stored_value(prisma) -> str:
     """Pull the credential_b64 value passed to the most recent upsert call."""
     call = prisma.db.litellm_mcpusercredentials.upsert.call_args
@@ -136,9 +395,7 @@ async def test_store_user_oauth_credential_does_not_persist_plaintext():
     access_token = "ya29.a0AfH6SMBverysecretaccesstoken"
     prisma = _make_prisma_with_existing(row=None)
 
-    await store_user_oauth_credential(
-        prisma, "alice", "srv-1", access_token, refresh_token="rfr-xyz"
-    )
+    await store_user_oauth_credential(prisma, "alice", "srv-1", access_token, refresh_token="rfr-xyz")
 
     stored = _stored_value(prisma)
     try:
@@ -221,9 +478,7 @@ async def test_byok_guard_rejects_overwriting_encrypted_byok():
 
     encrypted_row = MagicMock()
     encrypted_row.credential_b64 = _stored_value(prisma)
-    prisma.db.litellm_mcpusercredentials.find_unique = AsyncMock(
-        return_value=encrypted_row
-    )
+    prisma.db.litellm_mcpusercredentials.find_unique = AsyncMock(return_value=encrypted_row)
 
     with pytest.raises(ValueError, match="could not be verified as an OAuth2"):
         await store_user_oauth_credential(prisma, "alice", "srv-1", "tok")
@@ -265,18 +520,14 @@ async def test_list_oauth_credentials_filters_byok_and_returns_payloads():
         "connected_at": "2024-01-01T00:00:00Z",
     }
     legacy_row = MagicMock()
-    legacy_row.credential_b64 = base64.urlsafe_b64encode(
-        json.dumps(legacy_payload).encode()
-    ).decode()
+    legacy_row.credential_b64 = base64.urlsafe_b64encode(json.dumps(legacy_payload).encode()).decode()
     legacy_row.server_id = "srv-legacy"
 
     byok_row = MagicMock()
     byok_row.credential_b64 = base64.urlsafe_b64encode(b"plain-byok-key").decode()
     byok_row.server_id = "srv-byok"
 
-    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(
-        return_value=[encrypted_row, legacy_row, byok_row]
-    )
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[encrypted_row, legacy_row, byok_row])
 
     results = await list_user_oauth_credentials(prisma, "alice")
 
@@ -326,9 +577,7 @@ async def test_rotate_re_encrypts_byok_with_new_key(monkeypatch):
     prisma.db.litellm_mcpusercredentials.update = AsyncMock()
 
     new_master_key = "rotated-salt-key-9999-9999-9999-9999"
-    await rotate_mcp_user_credentials_master_key(
-        prisma_client=prisma, new_master_key=new_master_key
-    )
+    await rotate_mcp_user_credentials_master_key(prisma_client=prisma, new_master_key=new_master_key)
 
     update_call = prisma.db.litellm_mcpusercredentials.update.call_args
     new_stored = update_call.kwargs["data"]["credential_b64"]
@@ -356,19 +605,13 @@ async def test_rotate_migrates_legacy_plaintext_rows(monkeypatch):
     legacy_row.user_id = "alice"
     legacy_row.server_id = "srv-legacy"
     legacy_row.credential_b64 = base64.urlsafe_b64encode(b"legacy-plain").decode()
-    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(
-        return_value=[legacy_row]
-    )
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[legacy_row])
     prisma.db.litellm_mcpusercredentials.update = AsyncMock()
 
     new_key = "another-rotation-key-aaaa-bbbb-cccc-dddd"
-    await rotate_mcp_user_credentials_master_key(
-        prisma_client=prisma, new_master_key=new_key
-    )
+    await rotate_mcp_user_credentials_master_key(prisma_client=prisma, new_master_key=new_key)
 
-    new_stored = prisma.db.litellm_mcpusercredentials.update.call_args.kwargs["data"][
-        "credential_b64"
-    ]
+    new_stored = prisma.db.litellm_mcpusercredentials.update.call_args.kwargs["data"]["credential_b64"]
     monkeypatch.setenv("LITELLM_SALT_KEY", new_key)
     assert (
         decrypt_value_helper(
@@ -396,14 +639,10 @@ async def test_rotate_skips_undecodable_rows():
     good_row.server_id = "srv-ok"
     good_row.credential_b64 = base64.urlsafe_b64encode(b"good-byok").decode()
 
-    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(
-        return_value=[bad_row, good_row]
-    )
+    prisma.db.litellm_mcpusercredentials.find_many = AsyncMock(return_value=[bad_row, good_row])
     prisma.db.litellm_mcpusercredentials.update = AsyncMock()
 
-    await rotate_mcp_user_credentials_master_key(
-        prisma_client=prisma, new_master_key="new-key-xxxx"
-    )
+    await rotate_mcp_user_credentials_master_key(prisma_client=prisma, new_master_key="new-key-xxxx")
 
     # Only one update call — the good row.
     assert prisma.db.litellm_mcpusercredentials.update.call_count == 1
@@ -419,9 +658,7 @@ def _oauth_cred(access_token="at-live", refresh_token=None, expires_in_seconds=N
     if refresh_token is not None:
         cred["refresh_token"] = refresh_token
     if expires_in_seconds is not None:
-        cred["expires_at"] = (
-            datetime.now(timezone.utc) + timedelta(seconds=expires_in_seconds)
-        ).isoformat()
+        cred["expires_at"] = (datetime.now(timezone.utc) + timedelta(seconds=expires_in_seconds)).isoformat()
     return cred
 
 
@@ -438,12 +675,7 @@ def test_expiry_buffer_treats_soon_to_expire_as_expired():
     cred = _oauth_cred(expires_in_seconds=30)
     assert is_oauth_credential_expired(cred, buffer_seconds=60) is True
     # A token comfortably beyond the buffer stays valid.
-    assert (
-        is_oauth_credential_expired(
-            _oauth_cred(expires_in_seconds=600), buffer_seconds=60
-        )
-        is False
-    )
+    assert is_oauth_credential_expired(_oauth_cred(expires_in_seconds=600), buffer_seconds=60) is False
 
 
 def test_expiry_past_is_expired_regardless_of_buffer():
@@ -465,9 +697,7 @@ async def test_resolve_returns_valid_token_without_refreshing(monkeypatch):
     refresh = AsyncMock()
     monkeypatch.setattr(db_mod, "refresh_user_oauth_token", refresh)
 
-    cred = _oauth_cred(
-        access_token="at-live", refresh_token="rt-1", expires_in_seconds=600
-    )
+    cred = _oauth_cred(access_token="at-live", refresh_token="rt-1", expires_in_seconds=600)
     result = await resolve_valid_user_oauth_token(
         user_id="alice", server=MagicMock(), cred=cred, prisma_client=MagicMock()
     )
@@ -483,15 +713,11 @@ async def test_resolve_refreshes_expired_token_with_refresh_token(monkeypatch):
     # new token rather than returning None (which left the UI tool list empty).
     import litellm.proxy._experimental.mcp_server.db as db_mod
 
-    refreshed = _oauth_cred(
-        access_token="at-fresh", refresh_token="rt-2", expires_in_seconds=3600
-    )
+    refreshed = _oauth_cred(access_token="at-fresh", refresh_token="rt-2", expires_in_seconds=3600)
     refresh = AsyncMock(return_value=refreshed)
     monkeypatch.setattr(db_mod, "refresh_user_oauth_token", refresh)
 
-    expired = _oauth_cred(
-        access_token="at-dead", refresh_token="rt-1", expires_in_seconds=-5
-    )
+    expired = _oauth_cred(access_token="at-dead", refresh_token="rt-1", expires_in_seconds=-5)
     result = await resolve_valid_user_oauth_token(
         user_id="alice", server=MagicMock(), cred=expired, prisma_client=MagicMock()
     )
@@ -510,9 +736,7 @@ async def test_resolve_refreshes_token_expiring_within_buffer(monkeypatch):
     refresh = AsyncMock(return_value=refreshed)
     monkeypatch.setattr(db_mod, "refresh_user_oauth_token", refresh)
 
-    soon = _oauth_cred(
-        access_token="at-soon", refresh_token="rt-1", expires_in_seconds=30
-    )
+    soon = _oauth_cred(access_token="at-soon", refresh_token="rt-1", expires_in_seconds=30)
     result = await resolve_valid_user_oauth_token(
         user_id="alice", server=MagicMock(), cred=soon, prisma_client=MagicMock()
     )
@@ -546,9 +770,7 @@ async def test_resolve_returns_none_when_refresh_fails(monkeypatch):
     refresh = AsyncMock(return_value=None)
     monkeypatch.setattr(db_mod, "refresh_user_oauth_token", refresh)
 
-    expired = _oauth_cred(
-        access_token="at-dead", refresh_token="rt-1", expires_in_seconds=-5
-    )
+    expired = _oauth_cred(access_token="at-dead", refresh_token="rt-1", expires_in_seconds=-5)
     result = await resolve_valid_user_oauth_token(
         user_id="alice", server=MagicMock(), cred=expired, prisma_client=MagicMock()
     )
@@ -565,9 +787,7 @@ async def test_resolve_returns_none_for_missing_credential(monkeypatch):
     monkeypatch.setattr(db_mod, "refresh_user_oauth_token", refresh)
 
     assert (
-        await resolve_valid_user_oauth_token(
-            user_id="alice", server=MagicMock(), cred=None, prisma_client=MagicMock()
-        )
+        await resolve_valid_user_oauth_token(user_id="alice", server=MagicMock(), cred=None, prisma_client=MagicMock())
         is None
     )
     assert (
@@ -601,19 +821,13 @@ async def test_rotate_user_env_vars_re_encrypts_with_new_key(monkeypatch):
     encrypted_old = encrypt_value_helper(json.dumps(values))
 
     prisma = MagicMock()
-    prisma.db.litellm_mcpuserenvvars.find_many = AsyncMock(
-        return_value=[_env_var_row(encrypted_old)]
-    )
+    prisma.db.litellm_mcpuserenvvars.find_many = AsyncMock(return_value=[_env_var_row(encrypted_old)])
     prisma.db.litellm_mcpuserenvvars.update = AsyncMock()
 
     new_master_key = "rotated-env-key-1111-2222-3333-4444"
-    await rotate_mcp_user_env_vars_master_key(
-        prisma_client=prisma, new_master_key=new_master_key
-    )
+    await rotate_mcp_user_env_vars_master_key(prisma_client=prisma, new_master_key=new_master_key)
 
-    new_stored = prisma.db.litellm_mcpuserenvvars.update.call_args.kwargs["data"][
-        "values_b64"
-    ]
+    new_stored = prisma.db.litellm_mcpuserenvvars.update.call_args.kwargs["data"]["values_b64"]
     assert new_stored != encrypted_old, "rotation must produce different ciphertext"
 
     monkeypatch.setenv("LITELLM_SALT_KEY", new_master_key)
@@ -630,18 +844,14 @@ async def test_rotate_user_env_vars_re_encrypts_with_new_key(monkeypatch):
 async def test_rotate_user_env_vars_skips_undecryptable_rows():
     # A corrupt row must be skipped (not overwritten) so recoverable data is
     # preserved and one bad row does not abort the rest of the rotation.
-    good = _env_var_row(
-        encrypt_value_helper(json.dumps({"A": "1"})), server_id="srv-ok"
-    )
+    good = _env_var_row(encrypt_value_helper(json.dumps({"A": "1"})), server_id="srv-ok")
     bad = _env_var_row("!!! not encrypted !!!", server_id="srv-corrupt")
 
     prisma = MagicMock()
     prisma.db.litellm_mcpuserenvvars.find_many = AsyncMock(return_value=[bad, good])
     prisma.db.litellm_mcpuserenvvars.update = AsyncMock()
 
-    await rotate_mcp_user_env_vars_master_key(
-        prisma_client=prisma, new_master_key="new-key-xxxx"
-    )
+    await rotate_mcp_user_env_vars_master_key(prisma_client=prisma, new_master_key="new-key-xxxx")
 
     assert prisma.db.litellm_mcpuserenvvars.update.call_count == 1
     where = prisma.db.litellm_mcpuserenvvars.update.call_args.kwargs["where"]
@@ -669,9 +879,7 @@ async def test_refresh_user_oauth_token_uses_client_secret_basic(monkeypatch):
 
     monkeypatch.setattr(db_mod, "get_async_httpx_client", lambda **kwargs: mock_client)
     monkeypatch.setattr(db_mod, "store_user_oauth_credential", AsyncMock())
-    monkeypatch.setattr(
-        db_mod, "get_user_oauth_credential", AsyncMock(return_value={"access_token": "new-at"})
-    )
+    monkeypatch.setattr(db_mod, "get_user_oauth_credential", AsyncMock(return_value={"access_token": "new-at"}))
 
     result = await db_mod.refresh_user_oauth_token(
         prisma_client=MagicMock(),
@@ -710,9 +918,7 @@ async def test_refresh_user_oauth_token_defaults_to_client_secret_post(monkeypat
 
     monkeypatch.setattr(db_mod, "get_async_httpx_client", lambda **kwargs: mock_client)
     monkeypatch.setattr(db_mod, "store_user_oauth_credential", AsyncMock())
-    monkeypatch.setattr(
-        db_mod, "get_user_oauth_credential", AsyncMock(return_value={"access_token": "new-at"})
-    )
+    monkeypatch.setattr(db_mod, "get_user_oauth_credential", AsyncMock(return_value={"access_token": "new-at"}))
 
     await db_mod.refresh_user_oauth_token(
         prisma_client=MagicMock(),

tests/test_litellm/proxy/_experimental/mcp_server/test_discoverable_endpoints.py

@@ -665,6 +665,178 @@ async def test_register_client_persists_dcr_client_identity():
     mock_update_server.assert_called_once()
 
 
+async def _register_persistence_attempted_for_auth_type(auth_type: MCPAuth) -> bool:
+    """Run register_client_with_server with persist_credentials=True for a server of ``auth_type``
+    and report whether the DCR result was persisted onto the server row. The client-forwarded token
+    modes must skip the persist even on the admin path: writing it stamps oauth2_flow and a
+    client_id onto a server whose contract is that the gateway stores nothing, which makes a fresh
+    pass-through server read as gateway-authorized. The upstream registration must still be relayed
+    to the browser either way, since the caller needs the minted client to run its own flow."""
+    from fastapi import Request
+
+    from litellm.proxy._experimental.mcp_server.discoverable_endpoints import (
+        register_client_with_server,
+    )
+    from litellm.proxy._experimental.mcp_server.mcp_server_manager import (
+        global_mcp_server_manager,
+    )
+    from litellm.proxy._types import MCPTransport
+    from litellm.types.mcp_server.mcp_server_manager import MCPServer
+
+    server = MCPServer(
+        server_id="pt_server",
+        name="pt_server",
+        server_name="pt_server",
+        alias="pt_server",
+        transport=MCPTransport.http,
+        auth_type=auth_type,
+        client_id=None,
+        client_secret=None,
+        authorization_url="https://provider.example/oauth/authorize",
+        token_url="https://provider.example/oauth/token",
+        registration_url="https://provider.example/oauth/register",
+    )
+
+    mock_request = MagicMock(spec=Request)
+    mock_request.base_url = "https://proxy.litellm.example/"
+    mock_request.headers = {}
+
+    mock_response = MagicMock()
+    mock_response.json.return_value = {
+        "client_id": "generated-client",
+        "client_secret": "generated-secret",
+        "token_endpoint_auth_method": "none",
+    }
+    mock_response.raise_for_status = MagicMock()
+    mock_async_client = MagicMock()
+    mock_async_client.post = AsyncMock(return_value=mock_response)
+
+    mock_update = AsyncMock(return_value=MagicMock())
+
+    with (
+        patch(
+            "litellm.proxy._experimental.mcp_server.discoverable_endpoints.get_async_httpx_client",
+            return_value=mock_async_client,
+        ),
+        patch("litellm.proxy.utils.get_prisma_client_or_throw", return_value=MagicMock()),
+        patch("litellm.proxy._experimental.mcp_server.db.update_mcp_server", new=mock_update),
+        patch.object(global_mcp_server_manager, "update_server", new=AsyncMock()),
+    ):
+        response = await register_client_with_server(
+            request=mock_request,
+            mcp_server=server,
+            client_name="Litellm Proxy",
+            grant_types=["authorization_code", "refresh_token"],
+            response_types=["code"],
+            token_endpoint_auth_method="none",
+            persist_credentials=True,
+        )
+
+    assert json.loads(response.body.decode("utf-8")) == mock_response.json.return_value
+    return mock_update.await_count > 0
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("auth_type", [MCPAuth.true_passthrough, MCPAuth.oauth_delegate])
+async def test_register_client_does_not_persist_for_client_forwarded_modes(auth_type):
+    """The admin Authorize path (persist_credentials=True) must not write the DCR client onto a
+    true_passthrough / oauth_delegate server row: the browser still receives the registration, but
+    the gateway keeps no OAuth client identity for these modes."""
+    assert await _register_persistence_attempted_for_auth_type(auth_type) is False
+
+
+@pytest.mark.asyncio
+async def test_register_client_persist_discriminator_oauth2_persists():
+    """Guard the no-persist assertion above against vacuity: the same helper run against a genuine
+    oauth2 server DOES persist, so a regression that silently disables persistence everywhere (or a
+    helper that never reaches the persist) fails here instead of passing both."""
+    assert await _register_persistence_attempted_for_auth_type(MCPAuth.oauth2) is True
+
+
+@pytest.mark.asyncio
+async def test_register_client_persists_only_to_its_own_row_when_another_server_shares_the_url():
+    """A fresh server must mint and persist its OWN DCR client even when another server row with
+    the same upstream URL already holds one: both the reuse lookup and the persist are keyed by
+    server_id, never by URL, so OAuth client identity is not transferable between server entries.
+    If either side ever falls back to a URL match, this fails: the fresh server would skip the
+    upstream registration (adopting the sibling's client) or persist onto the wrong row."""
+    from fastapi import Request
+
+    from litellm.proxy._experimental.mcp_server.discoverable_endpoints import (
+        register_client_with_server,
+    )
+    from litellm.proxy._experimental.mcp_server.mcp_server_manager import (
+        global_mcp_server_manager,
+    )
+    from litellm.proxy._types import MCPTransport
+    from litellm.types.mcp_server.mcp_server_manager import MCPServer
+
+    shared_url = "https://provider.example/mcp"
+    fresh_server = MCPServer(
+        server_id="server-b",
+        name="server-b",
+        server_name="server-b",
+        alias="server-b",
+        transport=MCPTransport.http,
+        auth_type=MCPAuth.oauth2,
+        url=shared_url,
+        client_id=None,
+        client_secret=None,
+        authorization_url="https://provider.example/oauth/authorize",
+        token_url="https://provider.example/oauth/token",
+        registration_url="https://provider.example/oauth/register",
+    )
+    sibling_row_with_client = MagicMock(server_id="server-a", url=shared_url)
+    sibling_row_with_client.credentials = {"client_id": "client-a-do-not-adopt"}
+    own_row_without_client = MagicMock(server_id="server-b", url=shared_url)
+    own_row_without_client.credentials = {}
+    rows_by_server_id = {"server-a": sibling_row_with_client, "server-b": own_row_without_client}
+
+    mock_request = MagicMock(spec=Request)
+    mock_request.base_url = "https://proxy.litellm.example/"
+    mock_request.headers = {}
+
+    mock_response = MagicMock()
+    mock_response.json.return_value = {"client_id": "fresh-client-b", "token_endpoint_auth_method": "none"}
+    mock_response.raise_for_status = MagicMock()
+    mock_async_client = MagicMock()
+    mock_async_client.post = AsyncMock(return_value=mock_response)
+
+    mock_update = AsyncMock(return_value=MagicMock())
+
+    async def _get_row(prisma_client, server_id):
+        return rows_by_server_id.get(server_id)
+
+    with (
+        patch(
+            "litellm.proxy._experimental.mcp_server.discoverable_endpoints.get_async_httpx_client",
+            return_value=mock_async_client,
+        ),
+        patch("litellm.proxy.utils.get_prisma_client_or_throw", return_value=MagicMock()),
+        patch("litellm.proxy._experimental.mcp_server.db.get_mcp_server", new=AsyncMock(side_effect=_get_row)),
+        patch("litellm.proxy._experimental.mcp_server.db.update_mcp_server", new=mock_update),
+        patch.object(global_mcp_server_manager, "update_server", new=AsyncMock()),
+    ):
+        response = await register_client_with_server(
+            request=mock_request,
+            mcp_server=fresh_server,
+            client_name="Litellm Proxy",
+            grant_types=["authorization_code", "refresh_token"],
+            response_types=["code"],
+            token_endpoint_auth_method="none",
+            persist_credentials=True,
+        )
+
+    mock_async_client.post.assert_called_once()
+    body = json.loads(response.body.decode("utf-8"))
+    assert body["client_id"] == "fresh-client-b"
+
+    mock_update.assert_called_once()
+    update_data = mock_update.call_args.kwargs["data"]
+    assert update_data.server_id == "server-b"
+    assert update_data.credentials["client_id"] == "fresh-client-b"
+
+
 @pytest.mark.asyncio
 async def test_register_client_does_not_clobber_token_url_when_absent():
     """When the in-memory server has no token_url, the DCR persist must omit it from the

tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_oauth_passthrough.py

@@ -110,6 +110,42 @@ def test_is_oauth_passthrough_false_without_authorization_header():
     assert server.is_oauth_passthrough is False
 
 
+@pytest.mark.parametrize("auth_type", [MCPAuth.true_passthrough, MCPAuth.oauth_delegate])
+def test_is_dcr_bridge_true_for_flagged_client_forwarded_modes(auth_type):
+    server = MCPServer(
+        server_id="s1",
+        name="s1",
+        transport=MCPTransport.http,
+        auth_type=auth_type,
+        dcr_bridge=True,
+    )
+    assert server.is_dcr_bridge is True
+
+
+@pytest.mark.parametrize("auth_type", [MCPAuth.true_passthrough, MCPAuth.oauth_delegate])
+def test_is_dcr_bridge_false_when_flag_unset(auth_type):
+    server = MCPServer(
+        server_id="s1",
+        name="s1",
+        transport=MCPTransport.http,
+        auth_type=auth_type,
+    )
+    assert server.dcr_bridge is None
+    assert server.is_dcr_bridge is False
+
+
+@pytest.mark.parametrize("auth_type", [MCPAuth.oauth2, MCPAuth.none, MCPAuth.api_key, None])
+def test_is_dcr_bridge_false_for_non_client_forwarded_auth_types(auth_type):
+    server = MCPServer(
+        server_id="s1",
+        name="s1",
+        transport=MCPTransport.http,
+        auth_type=auth_type,
+        dcr_bridge=True,
+    )
+    assert server.is_dcr_bridge is False
+
+
 def test_is_oauth_passthrough_false_without_extra_headers():
     server = MCPServer(
         server_id="s1",

tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_partial_update.py

@@ -201,6 +201,7 @@ async def test_auth_type_switch_clears_stale_flow_scoped_fields():
         "token_url",
         "registration_url",
         "oauth2_flow",
+        "dcr_bridge",
         "token_exchange_endpoint",
         "audience",
         "subject_token_type",
@@ -225,6 +226,20 @@ async def test_auth_type_switch_keeps_explicitly_provided_flow_fields():
     assert data_dict["token_url"] is None
 
 
+@pytest.mark.asyncio
+async def test_auth_type_switch_to_client_forwarded_keeps_explicit_dcr_bridge():
+    data = UpdateMCPServerRequest(
+        server_id="my-test-server",
+        auth_type="true_passthrough",
+        dcr_bridge=True,
+    )
+
+    data_dict = await _run_update_with_existing(data, existing_auth_type="oauth2")
+
+    assert data_dict["dcr_bridge"] is True
+    assert data_dict["oauth2_flow"] is None
+
+
 @pytest.mark.asyncio
 async def test_auth_type_switch_back_to_oauth2_clears_token_exchange_fields():
     """The reverse switch must not leave token-exchange settings behind to
@@ -256,6 +271,7 @@ async def test_unchanged_auth_type_does_not_clear_flow_fields():
         "token_url",
         "registration_url",
         "oauth2_flow",
+        "dcr_bridge",
         "token_exchange_endpoint",
         "audience",
         "subject_token_type",

tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_server.py

@@ -6869,6 +6869,7 @@ async def test_call_tool_with_legacy_db_m2m_server_resolves_oauth2_flow():
         (None, None),
         ("", None),
         ("not a url", None),
+        ("http://[::1", None),
     ],
 )
 def test_redact_mcp_resource_url_strips_credentials(url, expected):

tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_server_manager.py

@@ -373,6 +373,66 @@ class TestMCPServerManager:
         server = next(iter(manager.config_mcp_servers.values()))
         assert server.oauth2_flow is None
 
+    def _client_forwarded_config(self, auth_type, **overrides):
+        base = {
+            "url": "https://example.com/mcp",
+            "transport": MCPTransport.http,
+            "auth_type": auth_type,
+        }
+        base.update(overrides)
+        return {"bridgeserver": base}
+
+    @pytest.mark.asyncio
+    async def test_load_servers_from_config_rejects_dcr_bridge_on_gateway_managed_auth_type(self):
+        manager = MCPServerManager()
+
+        with (
+            patch.object(manager, "_descovery_metadata", new=AsyncMock(return_value=None)),
+            pytest.raises(ValueError) as exc_info,
+        ):
+            await manager.load_servers_from_config(
+                self._oauth2_config(oauth2_flow="authorization_code", dcr_bridge=True)
+            )
+
+        assert "dcr_bridge is only supported" in str(exc_info.value)
+
+    @pytest.mark.asyncio
+    async def test_load_servers_from_config_rejects_non_boolean_dcr_bridge(self):
+        manager = MCPServerManager()
+
+        with (
+            patch.object(manager, "_descovery_metadata", new=AsyncMock(return_value=None)),
+            pytest.raises(ValueError) as exc_info,
+        ):
+            await manager.load_servers_from_config(
+                self._client_forwarded_config(MCPAuth.true_passthrough, dcr_bridge="yes")
+            )
+
+        assert "must be a boolean" in str(exc_info.value)
+
+    @pytest.mark.asyncio
+    @pytest.mark.parametrize("auth_type", [MCPAuth.true_passthrough, MCPAuth.oauth_delegate])
+    async def test_load_servers_from_config_accepts_dcr_bridge_on_client_forwarded_modes(self, auth_type):
+        manager = MCPServerManager()
+
+        with patch.object(manager, "_descovery_metadata", new=AsyncMock(return_value=None)):
+            await manager.load_servers_from_config(self._client_forwarded_config(auth_type, dcr_bridge=True))
+
+        server = next(iter(manager.config_mcp_servers.values()))
+        assert server.dcr_bridge is True
+        assert server.is_dcr_bridge is True
+
+    @pytest.mark.asyncio
+    async def test_load_servers_from_config_dcr_bridge_defaults_off(self):
+        manager = MCPServerManager()
+
+        with patch.object(manager, "_descovery_metadata", new=AsyncMock(return_value=None)):
+            await manager.load_servers_from_config(self._client_forwarded_config(MCPAuth.true_passthrough))
+
+        server = next(iter(manager.config_mcp_servers.values()))
+        assert server.dcr_bridge is None
+        assert server.is_dcr_bridge is False
+
     @pytest.mark.asyncio
     async def test_load_servers_from_config_coerces_cost_string_to_float(self):
         """YAML 1.1 parses `7e-05` as a string; ingest must coerce it to float."""
@@ -3327,9 +3387,35 @@ class TestMCPServerManager:
         await manager.invalidate_user_oauth_token_cache("alice", "srv-1")
         assert store.invalidations == [("alice", "srv-1")]
 
+    @pytest.mark.asyncio
+    async def test_invalidate_user_oauth_token_cache_drops_legacy_cache_too(self):
+        """A per-user token can be served from the legacy per-user token cache as well as the v2
+        store; the shared invalidation must evict both, or the path not evicted keeps serving a
+        token minted for a replaced credential row until its TTL."""
+
+        class _Store:
+            async def fetch(self, user_id: str, server_id: str):
+                return None
+
+            async def invalidate(self, user_id: str, server_id: str) -> None:
+                return None
+
+        class _LegacyCache:
+            def __init__(self) -> None:
+                self.deletes: list[tuple[str, str]] = []
+
+            async def delete(self, user_id: str, server_id: str) -> None:
+                self.deletes.append((user_id, server_id))
+
+        legacy_cache = _LegacyCache()
+        manager = MCPServerManager(per_user_oauth_token_store=_Store(), per_user_token_cache=legacy_cache)
+        await manager.invalidate_user_oauth_token_cache("alice", "srv-1")
+        assert legacy_cache.deletes == [("alice", "srv-1")]
+
     @pytest.mark.asyncio
     async def test_invalidate_user_oauth_token_cache_swallows_store_errors(self):
-        """A cache-drop failure must not fail the credential write that triggered it."""
+        """A cache-drop failure must not fail the credential write that triggered it, and the
+        legacy cache must still be evicted after the v2 store drop fails."""
 
         class _Store:
             async def fetch(self, user_id: str, server_id: str):
@@ -3338,7 +3424,35 @@ class TestMCPServerManager:
             async def invalidate(self, user_id: str, server_id: str) -> None:
                 raise RuntimeError("redis down")
 
-        manager = MCPServerManager(per_user_oauth_token_store=_Store())
+        class _LegacyCache:
+            def __init__(self) -> None:
+                self.deletes: list[tuple[str, str]] = []
+
+            async def delete(self, user_id: str, server_id: str) -> None:
+                self.deletes.append((user_id, server_id))
+
+        legacy_cache = _LegacyCache()
+        manager = MCPServerManager(per_user_oauth_token_store=_Store(), per_user_token_cache=legacy_cache)
+        await manager.invalidate_user_oauth_token_cache("alice", "srv-1")
+        assert legacy_cache.deletes == [("alice", "srv-1")]
+
+    @pytest.mark.asyncio
+    async def test_invalidate_user_oauth_token_cache_swallows_legacy_cache_errors(self):
+        """The legacy cache drop is best-effort like the v2 drop: a failure must be logged, never
+        raised into the credential write that triggered the invalidation."""
+
+        class _Store:
+            async def fetch(self, user_id: str, server_id: str):
+                return None
+
+            async def invalidate(self, user_id: str, server_id: str) -> None:
+                return None
+
+        class _RaisingLegacyCache:
+            async def delete(self, user_id: str, server_id: str) -> None:
+                raise RuntimeError("redis down")
+
+        manager = MCPServerManager(per_user_oauth_token_store=_Store(), per_user_token_cache=_RaisingLegacyCache())
         await manager.invalidate_user_oauth_token_cache("alice", "srv-1")
 
     @pytest.mark.asyncio

tests/test_litellm/proxy/_experimental/mcp_server/test_oauth2_token_cache.py

@@ -92,6 +92,36 @@ async def test_token_cached_across_calls():
     assert mock_client.post.call_count == 1
 
 
+@pytest.mark.asyncio
+async def test_m2m_token_not_shared_across_server_ids_with_identical_config():
+    """Two servers with byte-identical client_credentials config but different server_ids must not
+    share a cached M2M token: the cache is keyed by server_id, so a new server entry (even one
+    recreated with the same URL and credentials) mints its own token instead of inheriting the
+    sibling's. Guards against the cache key ever collapsing to the URL or the client config."""
+    cache = MCPOAuth2TokenCache()
+    server_a = _server(server_id="srv-a")
+    server_b = _server(server_id="srv-b")
+    mock_client = AsyncMock()
+    mock_client.post.side_effect = [_token_response("tok-for-a"), _token_response("tok-for-b")]
+
+    with (
+        patch(
+            "litellm.proxy._experimental.mcp_server.oauth2_token_cache.get_async_httpx_client",
+            return_value=mock_client,
+        ),
+        patch(
+            "litellm.proxy._experimental.mcp_server.oauth2_token_cache.mcp_oauth2_token_cache",
+            cache,
+        ),
+    ):
+        token_a = await resolve_mcp_auth(server_a)
+        token_b = await resolve_mcp_auth(server_b)
+
+    assert token_a == "tok-for-a"
+    assert token_b == "tok-for-b"
+    assert mock_client.post.call_count == 2
+
+
 @pytest.mark.asyncio
 async def test_per_request_header_beats_oauth2():
     """An explicit mcp_auth_header takes priority over the OAuth2 token."""

tests/test_litellm/proxy/auth/test_route_checks.py

@@ -2595,6 +2595,135 @@ def test_org_admin_of_multiple_orgs_can_operate_on_both():
     assert _user_is_org_admin({"organizations": ["org-A", "org-B"]}, user_obj) is True
 
 
+# ── LIT-4221: /team/update org-context resolution from team_id ────────────────
+from litellm.proxy.auth.auth_checks_organization import (
+    add_team_org_context_to_request_body,
+)
+
+
+@pytest.mark.asyncio
+async def test_add_team_org_context_resolves_org_from_team():
+    """For /team/update with only team_id, the target team's org is resolved and
+    injected so the org-admin route gate can see it. This is what lets an org
+    admin update a team budget from the Hub UI, which sends team_id, not
+    organization_id (LIT-4221)."""
+
+    async def fetch(team_id: str):
+        assert team_id == "team-1"
+        return "org-1"
+
+    out = await add_team_org_context_to_request_body(
+        route="/team/update",
+        request_body={"team_id": "team-1", "max_budget": 42},
+        fetch_team_org_id=fetch,
+    )
+    assert out == {"team_id": "team-1", "max_budget": 42, "organization_id": "org-1"}
+
+
+@pytest.mark.asyncio
+async def test_add_team_org_context_noop_when_org_id_already_present():
+    """If the caller already passed organization_id, no lookup happens and the
+    body is returned unchanged."""
+
+    async def fetch(team_id: str):
+        raise AssertionError("must not resolve when organization_id is present")
+
+    body = {"team_id": "team-1", "organization_id": "org-explicit"}
+    out = await add_team_org_context_to_request_body(
+        route="/team/update", request_body=body, fetch_team_org_id=fetch
+    )
+    assert out == body
+
+
+@pytest.mark.asyncio
+async def test_add_team_org_context_noop_for_other_routes():
+    """Only /team/update opts into org resolution; other routes are untouched."""
+
+    async def fetch(team_id: str):
+        raise AssertionError("must not resolve for a non-opted-in route")
+
+    body = {"team_id": "team-1"}
+    out = await add_team_org_context_to_request_body(
+        route="/team/delete", request_body=body, fetch_team_org_id=fetch
+    )
+    assert out == body
+
+
+@pytest.mark.asyncio
+async def test_add_team_org_context_noop_when_team_has_no_org():
+    """A standalone team (no org) resolves to None, so nothing is injected and
+    the org-admin branch stays unreachable (no blanket access)."""
+
+    async def fetch(team_id: str):
+        return None
+
+    body = {"team_id": "team-1"}
+    out = await add_team_org_context_to_request_body(
+        route="/team/update", request_body=body, fetch_team_org_id=fetch
+    )
+    assert out == body
+
+
+def test_team_update_gate_allows_org_admin_with_resolved_org():
+    """Post-resolution (organization_id present), an org admin of that org clears
+    the gate for /team/update."""
+    user_obj = _make_org_admin_user("org-1")
+    valid_token = UserAPIKeyAuth(user_id="org-admin-user", user_role=LitellmUserRoles.INTERNAL_USER.value)
+    request = MagicMock(spec=Request)
+    request.method = "POST"
+    request.query_params = {}
+
+    RouteChecks.non_proxy_admin_allowed_routes_check(
+        user_obj=user_obj,
+        _user_role=LitellmUserRoles.INTERNAL_USER.value,
+        route="/team/update",
+        request=request,
+        valid_token=valid_token,
+        request_data={"team_id": "team-1", "organization_id": "org-1"},
+    )
+
+
+def test_team_update_gate_rejects_without_org_context():
+    """Without organization_id (i.e. resolution found no org, or a non-org-admin),
+    the gate still rejects /team/update — the fix adds no blanket allow. Guards
+    against re-widening the route (e.g. dropping it into self_managed_routes)."""
+    user_obj = _make_org_admin_user("org-1")
+    valid_token = UserAPIKeyAuth(user_id="org-admin-user", user_role=LitellmUserRoles.INTERNAL_USER.value)
+    request = MagicMock(spec=Request)
+    request.method = "POST"
+    request.query_params = {}
+
+    with pytest.raises(Exception):
+        RouteChecks.non_proxy_admin_allowed_routes_check(
+            user_obj=user_obj,
+            _user_role=LitellmUserRoles.INTERNAL_USER.value,
+            route="/team/update",
+            request=request,
+            valid_token=valid_token,
+            request_data={"team_id": "team-1", "max_budget": 42},
+        )
+
+
+def test_team_update_gate_rejects_cross_org_admin_with_resolved_org():
+    """Even after the target team's org is resolved, an org admin of a DIFFERENT
+    org is rejected at the gate (no cross-org escalation)."""
+    user_obj = _make_org_admin_user("org-1")
+    valid_token = UserAPIKeyAuth(user_id="org-admin-user", user_role=LitellmUserRoles.INTERNAL_USER.value)
+    request = MagicMock(spec=Request)
+    request.method = "POST"
+    request.query_params = {}
+
+    with pytest.raises(Exception):
+        RouteChecks.non_proxy_admin_allowed_routes_check(
+            user_obj=user_obj,
+            _user_role=LitellmUserRoles.INTERNAL_USER.value,
+            route="/team/update",
+            request=request,
+            valid_token=valid_token,
+            request_data={"team_id": "team-1", "organization_id": "org-2"},
+        )
+
+
 @pytest.mark.asyncio
 async def test_initialize_pass_through_registers_wildcard_for_auth_subpath():
     """

tests/test_litellm/proxy/management_endpoints/test_mcp_management_endpoints.py

@@ -4900,6 +4900,158 @@ def test_oauth2_flow_defaults_to_none_when_omitted():
     assert LiteLLM_MCPServerTable(server_id="srv-1", transport="http").oauth2_flow is None
 
 
+def test_dcr_bridge_rejected_on_create_for_gateway_managed_auth_type():
+    from pydantic import ValidationError
+
+    from litellm.proxy._types import NewMCPServerRequest
+
+    with pytest.raises(ValidationError) as exc:
+        NewMCPServerRequest(
+            server_name="bridge-server",
+            url="https://example.com/mcp",
+            transport="http",
+            auth_type="oauth2",
+            oauth2_flow="authorization_code",
+            dcr_bridge=True,
+        )
+    assert "dcr_bridge is only supported" in str(exc.value)
+
+
+def test_dcr_bridge_rejected_on_create_when_auth_type_omitted():
+    from pydantic import ValidationError
+
+    from litellm.proxy._types import NewMCPServerRequest
+
+    with pytest.raises(ValidationError) as exc:
+        NewMCPServerRequest(
+            server_name="bridge-server",
+            url="https://example.com/mcp",
+            transport="http",
+            dcr_bridge=True,
+        )
+    assert "dcr_bridge is only supported" in str(exc.value)
+
+
+@pytest.mark.parametrize("auth_type", ["true_passthrough", "oauth_delegate"])
+def test_dcr_bridge_accepted_on_create_for_client_forwarded_modes(auth_type):
+    from litellm.proxy._experimental.mcp_server.db import _prepare_mcp_server_data
+    from litellm.proxy._types import NewMCPServerRequest
+
+    payload = NewMCPServerRequest(
+        server_name="bridge-server",
+        url="https://example.com/mcp",
+        transport="http",
+        auth_type=auth_type,
+        dcr_bridge=True,
+    )
+    data_dict = _prepare_mcp_server_data(payload)
+    assert data_dict["dcr_bridge"] is True
+
+
+def test_dcr_bridge_update_rejected_when_payload_auth_type_not_client_forwarded():
+    from pydantic import ValidationError
+
+    from litellm.proxy._types import UpdateMCPServerRequest
+
+    with pytest.raises(ValidationError) as exc:
+        UpdateMCPServerRequest(server_id="srv-1", auth_type="oauth2", dcr_bridge=True)
+    assert "dcr_bridge is only supported" in str(exc.value)
+
+
+def test_dcr_bridge_update_without_auth_type_defers_to_endpoint():
+    from litellm.proxy._types import UpdateMCPServerRequest
+
+    assert UpdateMCPServerRequest(server_id="srv-1", dcr_bridge=True).dcr_bridge is True
+
+
+def test_dcr_bridge_round_trips_on_response_model():
+    from litellm.proxy._types import LiteLLM_MCPServerTable
+
+    row = LiteLLM_MCPServerTable(server_id="srv-1", transport="http", dcr_bridge=True)
+    assert row.dcr_bridge is True
+    assert LiteLLM_MCPServerTable(server_id="srv-1", transport="http").dcr_bridge is None
+
+
+def _edit_endpoint_patches(old_record, update_mock):
+    return (
+        patch("litellm.proxy.management_endpoints.mcp_management_endpoints.MCP_AVAILABLE", True),
+        patch(
+            "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw",
+            return_value=MagicMock(),
+        ),
+        patch(
+            "litellm.proxy.management_endpoints.mcp_management_endpoints.get_mcp_server",
+            AsyncMock(side_effect=old_record) if isinstance(old_record, Exception) else AsyncMock(return_value=old_record),
+        ),
+        patch(
+            "litellm.proxy.management_endpoints.mcp_management_endpoints.update_mcp_server",
+            update_mock,
+        ),
+        patch(
+            "litellm.proxy.management_endpoints.mcp_management_endpoints.validate_and_normalize_mcp_server_payload",
+            autospec=True,
+        ),
+    )
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("stored_auth_type", ["oauth2", "api_key", "none"])
+async def test_edit_mcp_server_rejects_dcr_bridge_when_stored_auth_type_not_client_forwarded(stored_auth_type):
+    from litellm.proxy._types import UpdateMCPServerRequest
+    from litellm.proxy.management_endpoints.mcp_management_endpoints import edit_mcp_server
+
+    old_record = MagicMock()
+    old_record.auth_type = stored_auth_type
+    update_mock = AsyncMock()
+    p1, p2, p3, p4, p5 = _edit_endpoint_patches(old_record, update_mock)
+    with p1, p2, p3, p4, p5:
+        payload = UpdateMCPServerRequest(server_id="srv-1", dcr_bridge=True)
+        user_auth = UserAPIKeyAuth(user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN)
+        with pytest.raises(HTTPException) as exc:
+            await edit_mcp_server(payload=payload, user_api_key_dict=user_auth)
+
+    assert exc.value.status_code == 400
+    assert "dcr_bridge is only supported" in str(exc.value.detail)
+    update_mock.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_edit_mcp_server_rejects_dcr_bridge_when_stored_record_unreadable():
+    from litellm.proxy._types import UpdateMCPServerRequest
+    from litellm.proxy.management_endpoints.mcp_management_endpoints import edit_mcp_server
+
+    update_mock = AsyncMock()
+    p1, p2, p3, p4, p5 = _edit_endpoint_patches(RuntimeError("db down"), update_mock)
+    with p1, p2, p3, p4, p5:
+        payload = UpdateMCPServerRequest(server_id="srv-1", dcr_bridge=True)
+        user_auth = UserAPIKeyAuth(user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN)
+        with pytest.raises(HTTPException) as exc:
+            await edit_mcp_server(payload=payload, user_api_key_dict=user_auth)
+
+    assert exc.value.status_code == 400
+    update_mock.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_edit_mcp_server_dcr_bridge_on_unknown_server_returns_404_not_400():
+    """A dcr_bridge enablement targeting a server_id that does not exist must surface the accurate
+    404 from the update path, not a misleading 400 about the stored auth_type: get_mcp_server
+    returns None for a missing row without raising, which is distinct from a failed read."""
+    from litellm.proxy._types import UpdateMCPServerRequest
+    from litellm.proxy.management_endpoints.mcp_management_endpoints import edit_mcp_server
+
+    update_mock = AsyncMock(return_value=None)
+    p1, p2, p3, p4, p5 = _edit_endpoint_patches(None, update_mock)
+    with p1, p2, p3, p4, p5:
+        payload = UpdateMCPServerRequest(server_id="does-not-exist", dcr_bridge=True)
+        user_auth = UserAPIKeyAuth(user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN)
+        with pytest.raises(HTTPException) as exc:
+            await edit_mcp_server(payload=payload, user_api_key_dict=user_auth)
+
+    assert exc.value.status_code == 404
+    update_mock.assert_called_once()
+
+
 class TestPerUserCredentialConfigServerResolution:
     """Per-user credential and env-var endpoints must resolve config-defined MCP
     servers, which live only in the in-memory registry and never get a DB row, so
@@ -5134,3 +5286,93 @@ def test_stamp_oauth2_flow_ignores_non_oauth2():
     payload = _oauth2_create_payload(auth_type="none")
     mgmt_endpoints.stamp_omitted_oauth2_flow(payload)
     assert payload.oauth2_flow is None
+
+
+async def _run_edit(old_record, updated_record, purge_mock=None):
+    from litellm.proxy.management_endpoints.mcp_management_endpoints import edit_mcp_server
+
+    server_id = updated_record.server_id
+    with (
+        patch("litellm.proxy.management_endpoints.mcp_management_endpoints.MCP_AVAILABLE", True),
+        patch(
+            "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw",
+            return_value=MagicMock(),
+        ),
+        patch(
+            "litellm.proxy.management_endpoints.mcp_management_endpoints.get_mcp_server",
+            AsyncMock(side_effect=old_record)
+            if isinstance(old_record, Exception)
+            else AsyncMock(return_value=old_record),
+        ),
+        patch(
+            "litellm.proxy.management_endpoints.mcp_management_endpoints.update_mcp_server",
+            AsyncMock(return_value=updated_record),
+        ),
+        patch(
+            "litellm.proxy.management_endpoints.mcp_management_endpoints.validate_and_normalize_mcp_server_payload",
+            autospec=True,
+        ),
+        patch("litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager") as mock_manager,
+        patch(
+            "litellm.proxy.management_endpoints.mcp_management_endpoints.purge_user_oauth_credentials_for_server",
+            purge_mock if purge_mock is not None else AsyncMock(return_value=1),
+        ) as mock_purge,
+    ):
+        mock_manager.update_server = AsyncMock()
+        mock_manager.reload_servers_from_database = AsyncMock()
+        payload = UpdateMCPServerRequest(server_id=server_id, alias=updated_record.alias, url=updated_record.url)
+        user_auth = UserAPIKeyAuth(user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN)
+        result = await edit_mcp_server(payload=payload, user_api_key_dict=user_auth)
+        return result, mock_purge
+
+
+@pytest.mark.asyncio
+async def test_edit_mcp_server_purges_user_tokens_on_mint_relevant_change():
+    server_id = str(uuid.uuid4())
+    old = generate_mock_mcp_server_db_record(server_id=server_id, url="https://old.example.com/mcp")
+    updated = generate_mock_mcp_server_db_record(server_id=server_id, url="https://new.example.com/mcp")
+
+    result, mock_purge = await _run_edit(old, updated)
+
+    assert result.server_id == server_id
+    mock_purge.assert_awaited_once()
+    assert mock_purge.await_args.args[1] == server_id
+
+
+@pytest.mark.asyncio
+async def test_edit_mcp_server_skips_purge_when_identity_unchanged():
+    server_id = str(uuid.uuid4())
+    old = generate_mock_mcp_server_db_record(server_id=server_id, alias="Before")
+    updated = generate_mock_mcp_server_db_record(server_id=server_id, alias="After")
+
+    result, mock_purge = await _run_edit(old, updated)
+
+    assert result.server_id == server_id
+    mock_purge.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_edit_mcp_server_purge_failure_does_not_fail_the_edit():
+    """The purge is best-effort: a purge exception after a successful update must be swallowed and
+    logged, never turned into an error response for an edit whose primary job already succeeded."""
+    server_id = str(uuid.uuid4())
+    old = generate_mock_mcp_server_db_record(server_id=server_id, url="https://old.example.com/mcp")
+    updated = generate_mock_mcp_server_db_record(server_id=server_id, url="https://new.example.com/mcp")
+
+    result, mock_purge = await _run_edit(old, updated, purge_mock=AsyncMock(side_effect=RuntimeError("db down")))
+
+    assert result.server_id == server_id
+    mock_purge.assert_awaited_once()
+
+
+@pytest.mark.asyncio
+async def test_edit_mcp_server_snapshot_failure_skips_purge_but_edit_succeeds():
+    """The pre-update snapshot read is advisory (it only feeds the purge decision); a read failure
+    must skip the stale-token check with a warning, never fail the edit itself."""
+    server_id = str(uuid.uuid4())
+    updated = generate_mock_mcp_server_db_record(server_id=server_id, url="https://new.example.com/mcp")
+
+    result, mock_purge = await _run_edit(RuntimeError("db read failed"), updated)
+
+    assert result.server_id == server_id
+    mock_purge.assert_not_awaited()

tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py

@@ -3017,6 +3017,7 @@ async def test_build_ui_spend_logs_response_dict_rows_session_counts():
         return_value=[
             {
                 "session_id": session_id,
+                "session_total_spend": 15.0,
                 "mcp_tool_call_count": 1,
                 "mcp_tool_call_spend": 10.0,
             }
@@ -3044,6 +3045,10 @@ async def test_build_ui_spend_logs_response_dict_rows_session_counts():
     assert rows[1]["mcp_tool_call_count"] == 1
     assert rows[1]["mcp_tool_call_spend"] == 10.0
 
+    # Every row in the session carries the full session spend, not just its own
+    assert rows[0]["session_total_spend"] == 15.0
+    assert rows[1]["session_total_spend"] == 15.0
+
     # Row without a session_id defaults to 1
     assert rows[2]["session_total_count"] == 1
 
@@ -3055,6 +3060,64 @@ async def test_build_ui_spend_logs_response_dict_rows_session_counts():
     )
 
 
+@pytest.mark.asyncio
+async def test_build_ui_spend_logs_response_sums_multi_round_session_spend():
+    """
+    Regression test for LIT-4342: for a multi-round session the UI must show the
+    summed cost of every round, not just the first call.  _build_ui_spend_logs_response
+    enriches each row of a session with session_total_spend aggregated across the
+    whole session, scoped to the authorized api_keys of the page.
+    """
+    from litellm.proxy.spend_tracking.spend_management_endpoints import (
+        _build_ui_spend_logs_response,
+    )
+
+    session_id = "sess-multi-round"
+    api_key = "hashed-key-xyz"
+    # Three rounds of the same chat session with different per-call spend.
+    dict_rows = [
+        {"request_id": "req-1", "session_id": session_id, "call_type": "completion", "api_key": api_key, "spend": 0.01},
+        {"request_id": "req-2", "session_id": session_id, "call_type": "completion", "api_key": api_key, "spend": 0.02},
+        {"request_id": "req-3", "session_id": session_id, "call_type": "completion", "api_key": api_key, "spend": 0.03},
+    ]
+
+    mock_prisma = MagicMock()
+    mock_prisma.db.litellm_spendlogs.group_by = AsyncMock(
+        return_value=[{"session_id": session_id, "_count": {"session_id": 3}}]
+    )
+    # The raw aggregate query returns the full session spend (0.01 + 0.02 + 0.03).
+    mock_prisma.db.query_raw = AsyncMock(
+        return_value=[
+            {
+                "session_id": session_id,
+                "session_total_spend": 0.06,
+                "mcp_tool_call_count": 0,
+                "mcp_tool_call_spend": 0.0,
+            }
+        ]
+    )
+
+    result = await _build_ui_spend_logs_response(
+        prisma_client=mock_prisma,
+        data=dict_rows,
+        total_records=3,
+        page=1,
+        page_size=50,
+        total_pages=1,
+        enrich_session_counts=True,
+    )
+
+    rows = result["data"]
+    assert [row["session_total_spend"] for row in rows] == [0.06, 0.06, 0.06]
+    # No MCP calls in this session, so MCP fields must not be attached.
+    assert all("mcp_tool_call_count" not in row for row in rows)
+
+    # The aggregate must be scoped to the authorized api_keys of the page.
+    _, call_args, _ = mock_prisma.db.query_raw.mock_calls[0]
+    assert call_args[1] == [session_id]
+    assert call_args[2] == [api_key]
+
+
 # ---------------------------------------------------------------------------
 # Tests for /spend/logs team-member permission
 # ---------------------------------------------------------------------------

tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py

@@ -33,6 +33,7 @@ from litellm.proxy.spend_tracking.spend_tracking_utils import (
     _is_master_key,
     _redact_prompt_leaks_in_error_string,
     _sanitize_error_information_for_spend_logs,
+    _sanitize_guardrail_information_for_spend_logs,
     _sanitize_request_body_for_spend_logs_payload,
     _should_store_prompts_and_responses_in_spend_logs,
     get_logging_payload,
@@ -1263,6 +1264,260 @@ def test_get_spend_logs_metadata_guardrail_info_fallback_from_metadata():
     assert result["guardrail_information"] is None
 
 
+@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
+def test_sanitize_guardrail_information_redacts_all_prompt_carrying_fields_when_flag_false(
+    mock_should_store,
+):
+    """
+    match_details and classification are declared as structured metadata but
+    in-tree writers (litellm_content_filter, block_code_execution) inline
+    raw prompt content into them, so they leak the same way
+    guardrail_request/guardrail_response do. Redaction must cover all four.
+    """
+    mock_should_store.return_value = False
+    guardrail_info = [
+        {
+            "guardrail_name": "demo-echo-guard",
+            "guardrail_status": "success",
+            "guardrail_request": {"messages": [{"role": "user", "content": "hi"}]},
+            "guardrail_response": {"evaluated_input": "hi"},
+            "match_details": [{"type": "pattern", "snippet": "hi", "action_taken": "log"}],
+            "classification": {"intent": "x", "evidence": [{"match": "hi"}]},
+            "guardrail_action": "NONE",
+        }
+    ]
+
+    result = _sanitize_guardrail_information_for_spend_logs(guardrail_info)
+
+    assert result is not None
+    entry = result[0]
+    assert entry["guardrail_request"] == REDACTED_BY_LITELM_STRING
+    assert entry["guardrail_response"] == REDACTED_BY_LITELM_STRING
+    assert entry["match_details"] == REDACTED_BY_LITELM_STRING
+    assert entry["classification"] == REDACTED_BY_LITELM_STRING
+    assert entry["guardrail_name"] == "demo-echo-guard"
+    assert entry["guardrail_status"] == "success"
+    assert entry["guardrail_action"] == "NONE"
+
+
+@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
+def test_sanitize_guardrail_information_redacts_prompt_fields_when_flag_false(
+    mock_should_store,
+):
+    """
+    LIT-4314 Issue A regression: with store_prompts_in_spend_logs=False,
+    guardrail_request and guardrail_response must be redacted before they
+    land in LiteLLM_SpendLogs.metadata, while every other field on the
+    entry is preserved bit-for-bit.
+    """
+    mock_should_store.return_value = False
+    guardrail_info = [
+        {
+            "guardrail_name": "demo-echo-guard",
+            "guardrail_provider": "custom",
+            "guardrail_mode": "pre_call",
+            "guardrail_status": "success",
+            "guardrail_request": {
+                "messages": [{"role": "user", "content": "Say hi in 3 words"}],
+            },
+            "guardrail_response": {
+                "evaluated_input": "Say hi in 3 words",
+                "verdict": "allow",
+            },
+            "start_time": 1_700_000_000.0,
+            "end_time": 1_700_000_000.5,
+            "duration": 0.5,
+            "guardrail_id": "gd-42",
+            "masked_entity_count": {"EMAIL": 1},
+            "violation_categories": ["prompt_injection"],
+            "risk_score": 3.5,
+            "guardrail_action": "NONE",
+        }
+    ]
+
+    result = _sanitize_guardrail_information_for_spend_logs(guardrail_info)
+
+    assert result is not None
+    assert len(result) == 1
+    entry = result[0]
+    assert entry["guardrail_request"] == REDACTED_BY_LITELM_STRING
+    assert entry["guardrail_response"] == REDACTED_BY_LITELM_STRING
+    assert entry["guardrail_name"] == "demo-echo-guard"
+    assert entry["guardrail_provider"] == "custom"
+    assert entry["guardrail_mode"] == "pre_call"
+    assert entry["guardrail_status"] == "success"
+    assert entry["start_time"] == 1_700_000_000.0
+    assert entry["end_time"] == 1_700_000_000.5
+    assert entry["duration"] == 0.5
+    assert entry["guardrail_id"] == "gd-42"
+    assert entry["masked_entity_count"] == {"EMAIL": 1}
+    assert entry["violation_categories"] == ["prompt_injection"]
+    assert entry["risk_score"] == 3.5
+    assert entry["guardrail_action"] == "NONE"
+
+    assert guardrail_info[0]["guardrail_request"] == {
+        "messages": [{"role": "user", "content": "Say hi in 3 words"}],
+    }
+    assert guardrail_info[0]["guardrail_response"] == {
+        "evaluated_input": "Say hi in 3 words",
+        "verdict": "allow",
+    }
+
+
+@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
+def test_sanitize_guardrail_information_passthrough_when_flag_true(
+    mock_should_store,
+):
+    """
+    When store_prompts_in_spend_logs=True the sanitizer must be a no-op so
+    operators who explicitly opted in still see full guardrail payloads.
+    """
+    mock_should_store.return_value = True
+    guardrail_info = [
+        {
+            "guardrail_name": "content_filter",
+            "guardrail_status": "success",
+            "guardrail_request": {"messages": [{"role": "user", "content": "hi"}]},
+            "guardrail_response": {"verdict": "allow"},
+        }
+    ]
+
+    result = _sanitize_guardrail_information_for_spend_logs(guardrail_info)
+
+    assert result == guardrail_info
+
+
+@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
+def test_sanitize_guardrail_information_none_passthrough(mock_should_store):
+    mock_should_store.return_value = False
+    assert _sanitize_guardrail_information_for_spend_logs(None) is None
+
+
+@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
+def test_sanitize_guardrail_information_normalizes_bare_dict_input(mock_should_store):
+    """
+    Regression: xecguard (xecguard.py:246) assigns a bare dict to
+    standard_logging_object["guardrail_information"] even though the typed
+    contract is Optional[List[...]]. Without defensive normalization here,
+    the for-loop would iterate the dict's string keys and _redact...
+    would TypeError on {**"guardrail_name"}, taking down the entire
+    spend-log write via update_database's broad except.
+    """
+    mock_should_store.return_value = False
+    bare_dict_entry = {
+        "guardrail_name": "xecguard",
+        "guardrail_status": "success",
+        "guardrail_response": {"decision": "SAFE", "raw_prompt": "hi"},
+        "start_time": 1.0,
+        "end_time": 2.0,
+        "duration": 1.0,
+    }
+
+    result = _sanitize_guardrail_information_for_spend_logs(bare_dict_entry)
+
+    assert result is not None
+    assert isinstance(result, list)
+    assert len(result) == 1
+    entry = result[0]
+    assert entry["guardrail_response"] == REDACTED_BY_LITELM_STRING
+    assert entry["guardrail_name"] == "xecguard"
+    assert entry["guardrail_status"] == "success"
+    assert entry["start_time"] == 1.0
+
+
+@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
+def test_sanitize_guardrail_information_drops_non_dict_items_in_list(mock_should_store):
+    """
+    A stray non-dict item in the list (e.g. from a buggy caller that
+    accidentally appends a string) should be silently skipped instead of
+    crashing the spend-log write.
+    """
+    mock_should_store.return_value = False
+    mixed_input = [
+        {"guardrail_name": "x", "guardrail_response": {"leak": "hi"}},
+        "not-a-dict",
+        None,
+    ]
+
+    result = _sanitize_guardrail_information_for_spend_logs(mixed_input)
+
+    assert result == [{"guardrail_name": "x", "guardrail_response": REDACTED_BY_LITELM_STRING}]
+
+
+@patch("litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs")
+def test_sanitize_guardrail_information_preserves_absent_prompt_fields(mock_should_store):
+    """
+    Entries that never carried guardrail_request or guardrail_response must
+    not gain those keys after sanitization; consumers keying on presence
+    (`"guardrail_request" in entry`) would otherwise flip from absent to
+    the sentinel string.
+    """
+    mock_should_store.return_value = False
+    guardrail_info = [
+        {
+            "guardrail_name": "demo-echo-guard",
+            "guardrail_status": "success",
+            "guardrail_response": {"verdict": "allow", "evaluated_input": "hi"},
+        }
+    ]
+
+    result = _sanitize_guardrail_information_for_spend_logs(guardrail_info)
+
+    assert result is not None
+    entry = result[0]
+    assert "guardrail_request" not in entry
+    assert entry["guardrail_response"] == REDACTED_BY_LITELM_STRING
+    assert entry["guardrail_name"] == "demo-echo-guard"
+    assert entry["guardrail_status"] == "success"
+
+
+@patch("litellm.proxy.proxy_server.master_key", "sk-master")
+@patch(
+    "litellm.proxy.proxy_server.general_settings",
+    {"store_prompts_in_spend_logs": False},
+)
+def test_get_logging_payload_redacts_guardrail_prompt_fields_when_flag_false():
+    """
+    End-to-end wire-in check: get_logging_payload -> _get_spend_logs_metadata
+    -> sanitizer. Without the wire-in at line 139, the raw guardrail_response
+    lands in payload["metadata"] verbatim.
+    """
+    guardrail_info = [
+        {
+            "guardrail_name": "demo-echo-guard",
+            "guardrail_provider": "custom",
+            "guardrail_status": "success",
+            "guardrail_request": {"messages": [{"role": "user", "content": "secret"}]},
+            "guardrail_response": {"evaluated_input": "secret"},
+        }
+    ]
+    kwargs = {
+        "model": "gpt-4o-mini",
+        "litellm_call_id": "test-call-id",
+        "litellm_params": {
+            "metadata": {
+                "user_api_key": "test-key",
+                "standard_logging_guardrail_information": guardrail_info,
+            },
+            "proxy_server_request": {},
+        },
+    }
+
+    payload = get_logging_payload(
+        kwargs=kwargs,
+        response_obj={},
+        start_time=datetime.datetime.now(tz=timezone.utc),
+        end_time=datetime.datetime.now(tz=timezone.utc),
+    )
+
+    metadata_result = json.loads(payload["metadata"])
+    stored = metadata_result["guardrail_information"][0]
+    assert stored["guardrail_request"] == REDACTED_BY_LITELM_STRING
+    assert stored["guardrail_response"] == REDACTED_BY_LITELM_STRING
+    assert stored["guardrail_name"] == "demo-echo-guard"
+    assert stored["guardrail_status"] == "success"
+
+
 def test_get_logging_payload_guardrail_info_when_no_standard_logging_payload():
     """
     When a guardrail blocks a request before the LLM call, the standard_logging_object
@@ -1295,7 +1550,10 @@ def test_get_logging_payload_guardrail_info_when_no_standard_logging_payload():
     }
 
     with patch("litellm.proxy.proxy_server.master_key", "sk-master"):
-        with patch("litellm.proxy.proxy_server.general_settings", {}):
+        with patch(
+            "litellm.proxy.proxy_server.general_settings",
+            {"store_prompts_in_spend_logs": True},
+        ):
             payload = get_logging_payload(
                 kwargs=kwargs,
                 response_obj={},

tests/test_litellm/test_gpt_5_6_model_metadata.py

@@ -62,6 +62,77 @@ def test_openai_gpt_5_6_model_info(model):
     assert provider == "openai"
 
 
+AZURE_GLOBAL_MODELS = (
+    "azure/gpt-5.6",
+    "azure/gpt-5.6-sol",
+    "azure/gpt-5.6-terra",
+    "azure/gpt-5.6-luna",
+)
+
+AZURE_REGIONAL_MODELS = tuple(
+    f"azure/{region}/{tier}"
+    for region in ("us", "eu")
+    for tier in ("gpt-5.6", "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna")
+)
+
+
+def _tier_key(azure_model):
+    return azure_model.split("/")[-1]
+
+
+def _load_main():
+    json_path = Path(__file__).parents[2] / "model_prices_and_context_window.json"
+    with open(json_path) as f:
+        return json.load(f)
+
+
+@pytest.mark.parametrize("model", AZURE_GLOBAL_MODELS)
+def test_azure_gpt_5_6_global_model_info(model):
+    model_cost = _load_main()
+    info = model_cost.get(model)
+    assert info is not None, f"{model} not found in model_prices_and_context_window.json"
+
+    assert info["litellm_provider"] == "azure"
+    assert info["mode"] == "chat"
+
+    input_cost, output_cost, cache_read_cost, _ = STANDARD_PRICING[_tier_key(model)]
+    assert info["input_cost_per_token"] == input_cost
+    assert info["output_cost_per_token"] == output_cost
+    assert info["cache_read_input_token_cost"] == cache_read_cost
+
+    assert info["input_cost_per_token_above_272k_tokens"] == pytest.approx(input_cost * 2)
+    assert info["output_cost_per_token_above_272k_tokens"] == pytest.approx(output_cost * 1.5)
+    assert info["input_cost_per_token_priority"] == pytest.approx(input_cost * 2)
+    assert info["output_cost_per_token_priority"] == pytest.approx(output_cost * 2)
+    assert info["input_cost_per_token_above_272k_tokens_priority"] == pytest.approx(input_cost * 4)
+    assert info["output_cost_per_token_above_272k_tokens_priority"] == pytest.approx(output_cost * 3)
+
+    assert info["max_input_tokens"] == 1050000
+    assert info["max_output_tokens"] == 128000
+    assert info["supports_reasoning"] is True
+
+    routed_model, provider, _, _ = get_llm_provider(model=model)
+    assert provider == "azure"
+
+
+@pytest.mark.parametrize("model", AZURE_REGIONAL_MODELS)
+def test_azure_gpt_5_6_regional_model_info(model):
+    model_cost = _load_main()
+    info = model_cost.get(model)
+    assert info is not None, f"{model} not found in model_prices_and_context_window.json"
+
+    assert info["litellm_provider"] == "azure"
+    input_cost, output_cost, cache_read_cost, _ = STANDARD_PRICING[_tier_key(model)]
+
+    assert info["input_cost_per_token"] == pytest.approx(input_cost * 1.1)
+    assert info["output_cost_per_token"] == pytest.approx(output_cost * 1.1)
+    assert info["cache_read_input_token_cost"] == pytest.approx(cache_read_cost * 1.1)
+    assert info["input_cost_per_token_above_272k_tokens"] == pytest.approx(input_cost * 2.2)
+    assert info["output_cost_per_token_above_272k_tokens"] == pytest.approx(output_cost * 1.65)
+    assert info["input_cost_per_token_priority"] == pytest.approx(input_cost * 2.75)
+    assert info["output_cost_per_token_priority"] == pytest.approx(output_cost * 2.75)
+
+
 def test_gpt_5_6_backup_matches_main():
     """Ensure the bundled model cost map stays in sync with the canonical file."""
     repo_root = Path(__file__).parents[2]
@@ -73,7 +144,7 @@ def test_gpt_5_6_backup_matches_main():
     with open(backup_path) as f:
         backup_cost = json.load(f)
 
-    for model in GPT_5_6_MODELS:
+    for model in GPT_5_6_MODELS + AZURE_GLOBAL_MODELS + AZURE_REGIONAL_MODELS:
         assert backup_cost.get(model) == main_cost.get(model), (
             f"{model} differs between main and backup model cost maps"
         )

tests/test_litellm/test_muse_spark_1_1_model_metadata.py

@@ -0,0 +1,63 @@
+import json
+from pathlib import Path
+
+from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+MUSE_SPARK_MODEL = "meta/muse-spark-1.1"
+
+
+def test_muse_spark_1_1_model_info():
+    json_path = Path(__file__).parents[2] / "model_prices_and_context_window.json"
+    with open(json_path) as f:
+        model_cost = json.load(f)
+
+    info = model_cost.get(MUSE_SPARK_MODEL)
+    assert info is not None, f"{MUSE_SPARK_MODEL} not found in model_prices_and_context_window.json"
+
+    assert info["litellm_provider"] == "meta"
+    assert info["mode"] == "chat"
+
+    assert info["input_cost_per_token"] == 1.25e-06
+    assert info["output_cost_per_token"] == 4.25e-06
+    assert info["cache_read_input_token_cost"] == 1.5e-07
+
+    assert info["max_input_tokens"] == 1048576
+    assert info["max_output_tokens"] == 131072
+    assert info["max_tokens"] == 131072
+
+    assert info["supports_function_calling"] is True
+    assert info["supports_parallel_function_calling"] is True
+    assert info["supports_prompt_caching"] is True
+    assert info["supports_reasoning"] is True
+    assert info["supports_response_schema"] is True
+    assert info["supports_tool_choice"] is True
+    assert info["supports_vision"] is True
+    assert info["supports_pdf_input"] is True
+    assert info["supports_web_search"] is True
+    assert info["supports_minimal_reasoning_effort"] is True
+    assert info["supports_xhigh_reasoning_effort"] is True
+
+    assert info["supported_endpoints"] == ["/v1/chat/completions", "/v1/responses", "/v1/messages"]
+    assert info["supported_modalities"] == ["text", "image", "video"]
+    assert info["supported_output_modalities"] == ["text"]
+
+    routed_model, provider, _, api_base = get_llm_provider(model=MUSE_SPARK_MODEL, api_key="sk-test")
+    assert routed_model == "muse-spark-1.1"
+    assert provider == "meta"
+    assert api_base == "https://api.meta.ai/v1"
+
+
+def test_muse_spark_1_1_backup_matches_main():
+    """Ensure the bundled model cost map stays in sync with the canonical file."""
+    repo_root = Path(__file__).parents[2]
+    main_path = repo_root / "model_prices_and_context_window.json"
+    backup_path = repo_root / "litellm" / "model_prices_and_context_window_backup.json"
+
+    with open(main_path) as f:
+        main_cost = json.load(f)
+    with open(backup_path) as f:
+        backup_cost = json.load(f)
+
+    assert backup_cost.get(MUSE_SPARK_MODEL) == main_cost.get(MUSE_SPARK_MODEL), (
+        f"{MUSE_SPARK_MODEL} differs between main and backup model cost maps"
+    )

tests/test_litellm/test_utils.py

@@ -874,6 +874,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid():
                             "/v1/embeddings",
                             "/v1/chat/completions",
                             "/v1/completions",
+                            "/v1/messages",
                             "/v1/images/generations",
                             "/v1/realtime",
                             "/v1/realtime/transcription_sessions",
@@ -884,7 +885,6 @@ def test_aaamodel_prices_and_context_window_json_is_valid():
                             "/v1/audio/speech",
                             "/v1/ocr",
                             "/vertex_ai/live",
-                            "/v1/realtime/transcription_sessions",
                         ],
                     },
                 },

ui/litellm-dashboard/eslint-metrics.json

@@ -1,6 +1,6 @@
 {
-  "@typescript-eslint/no-explicit-any": 1980,
-  "complexity": 128,
+  "@typescript-eslint/no-explicit-any": 1977,
+  "complexity": 129,
   "local/no-large-inline-object-arg": 509,
   "local/no-long-condition-chain": 233,
   "max-depth": 59,

ui/litellm-dashboard/eslint-suppressions.json

@@ -4,32 +4,40 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/api-reference/APIReferenceView.tsx": {
+  "src/app/(dashboard)/admin-panel/_components/AdminPanel.tsx": {
+    "no-restricted-imports": {
+      "count": 1
+    },
+    "react-hooks/set-state-in-effect": {
+      "count": 1
+    }
+  },
+  "src/app/(dashboard)/api-reference/_components/APIReferenceView.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/budgets/components/budget_modal.tsx": {
+  "src/app/(dashboard)/budgets/_components/budget_modal.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/budgets/components/budget_panel.test.tsx": {
+  "src/app/(dashboard)/budgets/_components/budget_panel.test.tsx": {
     "unused-imports/no-unused-imports": {
       "count": 2
     }
   },
-  "src/app/(dashboard)/budgets/components/budget_panel.tsx": {
+  "src/app/(dashboard)/budgets/_components/budget_panel.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/budgets/components/edit_budget_modal.tsx": {
+  "src/app/(dashboard)/budgets/_components/edit_budget_modal.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/caching/components/cache_dashboard.tsx": {
+  "src/app/(dashboard)/caching/_components/cache_dashboard.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -40,17 +48,17 @@
       "count": 2
     }
   },
-  "src/app/(dashboard)/caching/components/cache_health.tsx": {
+  "src/app/(dashboard)/caching/_components/cache_health.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/caching/components/cache_settings/RedisTypeSelector.tsx": {
+  "src/app/(dashboard)/caching/_components/cache_settings/RedisTypeSelector.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/caching/components/cache_settings/index.tsx": {
+  "src/app/(dashboard)/caching/_components/cache_settings/index.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -58,17 +66,17 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/add_margin_form.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/add_margin_form.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/add_provider_form.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/add_provider_form.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/cost_tracking_settings.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/cost_tracking_settings.tsx": {
     "no-nested-ternary": {
       "count": 2
     },
@@ -76,96 +84,86 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/how_it_works.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/how_it_works.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/pricing_calculator/multi_cost_results.test.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/pricing_calculator/multi_cost_results.test.tsx": {
     "unused-imports/no-unused-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/pricing_calculator/multi_cost_results.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/pricing_calculator/multi_cost_results.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/pricing_calculator/multi_export_dropdown.test.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/pricing_calculator/multi_export_dropdown.test.tsx": {
     "unused-imports/no-unused-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/pricing_calculator/multi_export_dropdown.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/pricing_calculator/multi_export_dropdown.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/pricing_calculator/use_multi_cost_estimate.ts": {
+  "src/app/(dashboard)/cost-tracking/_components/pricing_calculator/use_multi_cost_estimate.ts": {
     "no-restricted-syntax": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/provider_discount_table.test.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/provider_discount_table.test.tsx": {
     "unused-imports/no-unused-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/provider_discount_table.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/provider_discount_table.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/provider_display_helpers.test.ts": {
+  "src/app/(dashboard)/cost-tracking/_components/provider_display_helpers.test.ts": {
     "unused-imports/no-unused-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/provider_margin_table.tsx": {
+  "src/app/(dashboard)/cost-tracking/_components/provider_margin_table.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/use_discount_config.ts": {
+  "src/app/(dashboard)/cost-tracking/_components/use_discount_config.ts": {
     "no-restricted-syntax": {
       "count": 2
     }
   },
-  "src/app/(dashboard)/cost-tracking/components/use_margin_config.ts": {
+  "src/app/(dashboard)/cost-tracking/_components/use_margin_config.ts": {
     "no-restricted-syntax": {
       "count": 2
     }
   },
-  "src/app/(dashboard)/guardrails-monitor/components/EvaluationSettingsModal.tsx": {
+  "src/app/(dashboard)/guardrails-monitor/_components/EvaluationSettingsModal.tsx": {
     "react-hooks/set-state-in-effect": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/guardrails-monitor/components/GuardrailDetail.tsx": {
+  "src/app/(dashboard)/guardrails-monitor/_components/GuardrailDetail.tsx": {
     "no-nested-ternary": {
       "count": 3
     }
   },
-  "src/app/(dashboard)/guardrails-monitor/components/GuardrailsMonitorView.tsx": {
+  "src/app/(dashboard)/guardrails-monitor/_components/GuardrailsMonitorView.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/guardrails-monitor/components/GuardrailsOverview.tsx": {
+  "src/app/(dashboard)/guardrails-monitor/_components/GuardrailsOverview.tsx": {
     "no-nested-ternary": {
       "count": 8
     }
   },
-  "src/app/(dashboard)/guardrails-monitor/components/ScoreChart.test.tsx": {
-    "react/display-name": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/guardrails-monitor/components/ScoreChart.tsx": {
-    "no-restricted-imports": {
-      "count": 1
-    }
-  },
   "src/app/(dashboard)/hooks/accessGroups/useAccessGroupDetails.ts": {
     "no-restricted-syntax": {
       "count": 1
@@ -326,7 +324,7 @@
       "count": 2
     }
   },
-  "src/app/(dashboard)/memory/components/MemoryView.tsx": {
+  "src/app/(dashboard)/memory/_components/MemoryView.tsx": {
     "react-hooks/set-state-in-effect": {
       "count": 1
     }
@@ -373,6 +371,22 @@
       "count": 1
     }
   },
+  "src/app/(dashboard)/old-usage/_components/usage.tsx": {
+    "no-restricted-imports": {
+      "count": 2
+    },
+    "react-hooks/immutability": {
+      "count": 1
+    },
+    "react-hooks/purity": {
+      "count": 1
+    }
+  },
+  "src/app/(dashboard)/organizations/_components/organizations.tsx": {
+    "no-restricted-imports": {
+      "count": 1
+    }
+  },
   "src/app/(dashboard)/playground/components/chat_ui/AdditionalModelSettings.tsx": {
     "no-restricted-imports": {
       "count": 1
@@ -522,7 +536,7 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/projects/components/ProjectDetailsPage.tsx": {
+  "src/app/(dashboard)/projects/_components/ProjectDetailsPage.tsx": {
     "no-nested-ternary": {
       "count": 3
     },
@@ -530,27 +544,27 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/projects/components/ProjectKeysSection.tsx": {
+  "src/app/(dashboard)/projects/_components/ProjectKeysSection.tsx": {
     "react-hooks/set-state-in-effect": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/projects/components/ProjectModals/ProjectBaseForm.tsx": {
+  "src/app/(dashboard)/projects/_components/ProjectModals/ProjectBaseForm.tsx": {
     "react-hooks/set-state-in-effect": {
       "count": 2
     }
   },
-  "src/app/(dashboard)/projects/components/ProjectsPage.tsx": {
+  "src/app/(dashboard)/projects/_components/ProjectsPage.tsx": {
     "react-hooks/set-state-in-effect": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/add_prompt_form.tsx": {
+  "src/app/(dashboard)/prompts/_components/add_prompt_form.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/index.tsx": {
+  "src/app/(dashboard)/prompts/_components/index.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -561,17 +575,17 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/DeveloperMessageCard.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/DeveloperMessageCard.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/ModelConfigCard.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/ModelConfigCard.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/PromptCodeSnippets.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/PromptCodeSnippets.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -582,32 +596,32 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/PromptEditorHeader.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/PromptEditorHeader.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/PromptMessagesCard.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/PromptMessagesCard.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/PublishModal.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/PublishModal.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/ToolsCard.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/ToolsCard.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/VersionHistorySidePanel.test.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/VersionHistorySidePanel.test.tsx": {
     "max-nested-callbacks": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/VersionHistorySidePanel.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/VersionHistorySidePanel.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -615,22 +629,22 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/conversation_panel/MessageInput.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/conversation_panel/MessageInput.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/conversation_panel/index.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/conversation_panel/index.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_editor_view/conversation_panel/useConversation.ts": {
+  "src/app/(dashboard)/prompts/_components/prompt_editor_view/conversation_panel/useConversation.ts": {
     "no-restricted-syntax": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_info.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_info.tsx": {
     "no-nested-ternary": {
       "count": 3
     },
@@ -641,7 +655,7 @@
       "count": 2
     }
   },
-  "src/app/(dashboard)/prompts/components/prompt_table.tsx": {
+  "src/app/(dashboard)/prompts/_components/prompt_table.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -649,6 +663,14 @@
       "count": 1
     }
   },
+  "src/app/(dashboard)/router-settings/_components/general_settings.tsx": {
+    "no-nested-ternary": {
+      "count": 3
+    },
+    "no-restricted-imports": {
+      "count": 2
+    }
+  },
   "src/app/(dashboard)/search-tools/_components/CreateSearchTools.tsx": {
     "no-restricted-imports": {
       "count": 1
@@ -851,14 +873,6 @@
       "count": 1
     }
   },
-  "src/components/AdminPanel.tsx": {
-    "no-restricted-imports": {
-      "count": 1
-    },
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
   "src/components/CreateUserButton.tsx": {
     "no-restricted-imports": {
       "count": 1
@@ -1247,7 +1261,7 @@
       "count": 1
     }
   },
-  "src/components/agents.tsx": {
+  "src/app/(dashboard)/agents/_components/index.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -1255,7 +1269,7 @@
       "count": 1
     }
   },
-  "src/components/agents/add_agent_form.tsx": {
+  "src/app/(dashboard)/agents/_components/add_agent_form.tsx": {
     "no-nested-ternary": {
       "count": 3
     },
@@ -1269,7 +1283,7 @@
       "count": 1
     }
   },
-  "src/components/agents/agent_card_discovery.tsx": {
+  "src/app/(dashboard)/agents/_components/agent_card_discovery.tsx": {
     "react-hooks/refs": {
       "count": 3
     },
@@ -1277,17 +1291,17 @@
       "count": 1
     }
   },
-  "src/components/agents/agent_cost_view.tsx": {
+  "src/app/(dashboard)/agents/_components/agent_cost_view.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/components/agents/agent_form_fields.tsx": {
+  "src/app/(dashboard)/agents/_components/agent_form_fields.tsx": {
     "no-nested-ternary": {
       "count": 1
     }
   },
-  "src/components/agents/agent_info.tsx": {
+  "src/app/(dashboard)/agents/_components/agent_info.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -1298,12 +1312,12 @@
       "count": 1
     }
   },
-  "src/components/agents/agent_virtual_keys.tsx": {
+  "src/app/(dashboard)/agents/_components/agent_virtual_keys.tsx": {
     "no-nested-ternary": {
       "count": 1
     }
   },
-  "src/components/agents/dynamic_agent_form_fields.tsx": {
+  "src/app/(dashboard)/agents/_components/dynamic_agent_form_fields.tsx": {
     "no-nested-ternary": {
       "count": 2
     }
@@ -1520,35 +1534,27 @@
       "count": 1
     }
   },
-  "src/components/general_settings.tsx": {
-    "no-nested-ternary": {
-      "count": 3
-    },
-    "no-restricted-imports": {
-      "count": 2
-    }
-  },
-  "src/components/guardrails.tsx": {
+  "src/app/(dashboard)/guardrails/_components/index.tsx": {
     "react-hooks/set-state-in-effect": {
       "count": 1
     }
   },
-  "src/components/guardrails/GuardrailTestPanel.tsx": {
+  "src/app/(dashboard)/guardrails/_components/GuardrailTestPanel.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/components/guardrails/GuardrailTestPlayground.tsx": {
+  "src/app/(dashboard)/guardrails/_components/GuardrailTestPlayground.tsx": {
     "no-nested-ternary": {
       "count": 1
     }
   },
-  "src/components/guardrails/GuardrailTestResults.tsx": {
+  "src/app/(dashboard)/guardrails/_components/GuardrailTestResults.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/components/guardrails/TeamGuardrailsTab.tsx": {
+  "src/app/(dashboard)/guardrails/_components/TeamGuardrailsTab.tsx": {
     "no-nested-ternary": {
       "count": 2
     },
@@ -1556,7 +1562,7 @@
       "count": 1
     }
   },
-  "src/components/guardrails/add_guardrail_form.tsx": {
+  "src/app/(dashboard)/guardrails/_components/add_guardrail_form.tsx": {
     "no-nested-ternary": {
       "count": 4
     },
@@ -1567,7 +1573,7 @@
       "count": 2
     }
   },
-  "src/components/guardrails/content_filter/CompetitorIntentConfiguration.tsx": {
+  "src/app/(dashboard)/guardrails/_components/content_filter/CompetitorIntentConfiguration.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -1575,7 +1581,7 @@
       "count": 1
     }
   },
-  "src/components/guardrails/content_filter/ContentCategoryConfiguration.tsx": {
+  "src/app/(dashboard)/guardrails/_components/content_filter/ContentCategoryConfiguration.tsx": {
     "no-nested-ternary": {
       "count": 3
     },
@@ -1583,12 +1589,12 @@
       "count": 1
     }
   },
-  "src/components/guardrails/content_filter/ContentFilterDisplay.tsx": {
+  "src/app/(dashboard)/guardrails/_components/content_filter/ContentFilterDisplay.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/components/guardrails/content_filter/ContentFilterManager.tsx": {
+  "src/app/(dashboard)/guardrails/_components/content_filter/ContentFilterManager.tsx": {
     "max-params": {
       "count": 2
     },
@@ -1596,7 +1602,7 @@
       "count": 1
     }
   },
-  "src/components/guardrails/custom_code/CustomCodeModal.tsx": {
+  "src/app/(dashboard)/guardrails/_components/custom_code/CustomCodeModal.tsx": {
     "no-nested-ternary": {
       "count": 6
     },
@@ -1607,7 +1613,7 @@
       "count": 1
     }
   },
-  "src/components/guardrails/edit_guardrail_form.tsx": {
+  "src/app/(dashboard)/guardrails/_components/edit_guardrail_form.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -1618,7 +1624,7 @@
       "count": 1
     }
   },
-  "src/components/guardrails/guardrail_info.tsx": {
+  "src/app/(dashboard)/guardrails/_components/guardrail_info.tsx": {
     "max-params": {
       "count": 1
     },
@@ -1629,7 +1635,7 @@
       "count": 3
     }
   },
-  "src/components/guardrails/guardrail_optional_params.tsx": {
+  "src/app/(dashboard)/guardrails/_components/guardrail_optional_params.tsx": {
     "no-nested-ternary": {
       "count": 5
     },
@@ -1637,7 +1643,7 @@
       "count": 1
     }
   },
-  "src/components/guardrails/guardrail_provider_fields.tsx": {
+  "src/app/(dashboard)/guardrails/_components/guardrail_provider_fields.tsx": {
     "no-nested-ternary": {
       "count": 5
     },
@@ -1645,7 +1651,7 @@
       "count": 1
     }
   },
-  "src/components/guardrails/guardrail_table.tsx": {
+  "src/app/(dashboard)/guardrails/_components/guardrail_table.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -1653,7 +1659,7 @@
       "count": 1
     }
   },
-  "src/components/guardrails/tool_permission/ToolPermissionRulesEditor.tsx": {
+  "src/app/(dashboard)/guardrails/_components/tool_permission/ToolPermissionRulesEditor.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2028,11 +2034,6 @@
       "count": 1
     }
   },
-  "src/components/organizations.tsx": {
-    "no-restricted-imports": {
-      "count": 1
-    }
-  },
   "src/components/page_utils.test.ts": {
     "max-nested-callbacks": {
       "count": 3
@@ -2079,7 +2080,7 @@
       "count": 1
     }
   },
-  "src/components/policies/add_attachment_form.tsx": {
+  "src/app/(dashboard)/policies/_components/add_attachment_form.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2087,7 +2088,7 @@
       "count": 1
     }
   },
-  "src/components/policies/add_policy_form.tsx": {
+  "src/app/(dashboard)/policies/_components/add_policy_form.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2098,7 +2099,7 @@
       "count": 1
     }
   },
-  "src/components/policies/ai_suggestion_modal.tsx": {
+  "src/app/(dashboard)/policies/_components/ai_suggestion_modal.tsx": {
     "no-nested-ternary": {
       "count": 10
     },
@@ -2109,12 +2110,12 @@
       "count": 1
     }
   },
-  "src/components/policies/attachment_table.test.tsx": {
+  "src/app/(dashboard)/policies/_components/attachment_table.test.tsx": {
     "react/display-name": {
       "count": 1
     }
   },
-  "src/components/policies/attachment_table.tsx": {
+  "src/app/(dashboard)/policies/_components/attachment_table.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -2122,7 +2123,7 @@
       "count": 1
     }
   },
-  "src/components/policies/guardrail_selection_modal.tsx": {
+  "src/app/(dashboard)/policies/_components/guardrail_selection_modal.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -2130,12 +2131,12 @@
       "count": 1
     }
   },
-  "src/components/policies/impact_popover.test.tsx": {
+  "src/app/(dashboard)/policies/_components/impact_popover.test.tsx": {
     "react/display-name": {
       "count": 1
     }
   },
-  "src/components/policies/impact_popover.tsx": {
+  "src/app/(dashboard)/policies/_components/impact_popover.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -2143,12 +2144,12 @@
       "count": 1
     }
   },
-  "src/components/policies/index.test.tsx": {
+  "src/app/(dashboard)/policies/_components/index.test.tsx": {
     "react/display-name": {
       "count": 1
     }
   },
-  "src/components/policies/index.tsx": {
+  "src/app/(dashboard)/policies/_components/index.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2156,7 +2157,7 @@
       "count": 1
     }
   },
-  "src/components/policies/pipeline_flow_builder.tsx": {
+  "src/app/(dashboard)/policies/_components/pipeline_flow_builder.tsx": {
     "no-nested-ternary": {
       "count": 1
     },
@@ -2167,7 +2168,7 @@
       "count": 2
     }
   },
-  "src/components/policies/policy_info.tsx": {
+  "src/app/(dashboard)/policies/_components/policy_info.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2175,12 +2176,12 @@
       "count": 1
     }
   },
-  "src/components/policies/policy_table.test.tsx": {
+  "src/app/(dashboard)/policies/_components/policy_table.test.tsx": {
     "react/display-name": {
       "count": 1
     }
   },
-  "src/components/policies/policy_table.tsx": {
+  "src/app/(dashboard)/policies/_components/policy_table.tsx": {
     "no-nested-ternary": {
       "count": 2
     },
@@ -2188,7 +2189,7 @@
       "count": 1
     }
   },
-  "src/components/policies/policy_test_panel.tsx": {
+  "src/app/(dashboard)/policies/_components/policy_test_panel.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2196,7 +2197,7 @@
       "count": 1
     }
   },
-  "src/components/policies/template_parameter_modal.tsx": {
+  "src/app/(dashboard)/policies/_components/template_parameter_modal.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2271,17 +2272,17 @@
       "count": 1
     }
   },
-  "src/components/tag_management/TagTable.tsx": {
+  "src/app/(dashboard)/tag-management/_components/TagTable.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/components/tag_management/components/CreateTagModal.tsx": {
+  "src/app/(dashboard)/tag-management/_components/components/CreateTagModal.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/components/tag_management/index.tsx": {
+  "src/app/(dashboard)/tag-management/_components/index.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2289,7 +2290,7 @@
       "count": 1
     }
   },
-  "src/components/tag_management/tag_info.tsx": {
+  "src/app/(dashboard)/tag-management/_components/tag_info.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2371,17 +2372,6 @@
       "count": 1
     }
   },
-  "src/components/usage.tsx": {
-    "no-restricted-imports": {
-      "count": 2
-    },
-    "react-hooks/immutability": {
-      "count": 1
-    },
-    "react-hooks/purity": {
-      "count": 1
-    }
-  },
   "src/components/user_agent_activity.tsx": {
     "no-restricted-imports": {
       "count": 2
@@ -2398,12 +2388,12 @@
       "count": 2
     }
   },
-  "src/components/vector_store_management/CreateVectorStore.tsx": {
+  "src/app/(dashboard)/vector-stores/_components/CreateVectorStore.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/components/vector_store_management/VectorStoreForm.tsx": {
+  "src/app/(dashboard)/vector-stores/_components/VectorStoreForm.tsx": {
     "no-nested-ternary": {
       "count": 2
     },
@@ -2414,12 +2404,12 @@
       "count": 1
     }
   },
-  "src/components/vector_store_management/VectorStoreTable.tsx": {
+  "src/app/(dashboard)/vector-stores/_components/VectorStoreTable.tsx": {
     "no-restricted-imports": {
       "count": 1
     }
   },
-  "src/components/vector_store_management/index.tsx": {
+  "src/app/(dashboard)/vector-stores/_components/index.tsx": {
     "no-restricted-imports": {
       "count": 1
     },
@@ -2427,7 +2417,7 @@
       "count": 1
     }
   },
-  "src/components/vector_store_management/vector_store_info.tsx": {
+  "src/app/(dashboard)/vector-stores/_components/vector_store_info.tsx": {
     "no-restricted-imports": {
       "count": 1
     },

ui/litellm-dashboard/package-lock.json

@@ -34,6 +34,7 @@
         "react-json-view-lite": "2.5.0",
         "react-markdown": "9.1.0",
         "react-syntax-highlighter": "15.6.6",
+        "recharts": "3.9.2",
         "remark-gfm": "4.0.1",
         "tailwind-merge": "3.4.0",
         "uuid": "14.0.0"
@@ -2927,6 +2928,32 @@
         "npm": ">=9.5.0"
       }
     },
+    "node_modules/@reduxjs/toolkit": {
+      "version": "2.12.0",
+      "resolved": "https://registry.npmjs.org/@reduxjs/toolkit/-/toolkit-2.12.0.tgz",
+      "integrity": "sha512-KiT+RzZbp6mQET+Mg+h2c97+9j1sNflUxQkIHI7Yuzf6Peu+OYpmkn6nbHWmLLWj+1ZODUJFwGZ7gx3L9R9EOw==",
+      "license": "MIT",
+      "dependencies": {
+        "@standard-schema/spec": "^1.0.0",
+        "@standard-schema/utils": "^0.3.0",
+        "immer": "^11.0.0",
+        "redux": "^5.0.1",
+        "redux-thunk": "^3.1.0",
+        "reselect": "^5.1.0"
+      },
+      "peerDependencies": {
+        "react": "^16.9.0 || ^17.0.0 || ^18 || ^19",
+        "react-redux": "^7.2.1 || ^8.1.3 || ^9.0.0"
+      },
+      "peerDependenciesMeta": {
+        "react": {
+          "optional": true
+        },
+        "react-redux": {
+          "optional": true
+        }
+      }
+    },
     "node_modules/@rollup/rollup-android-arm-eabi": {
       "version": "4.61.1",
       "resolved": "https://registry.npmjs.org/@rollup/rollup-android-arm-eabi/-/rollup-android-arm-eabi-4.61.1.tgz",
@@ -3284,6 +3311,18 @@
       "dev": true,
       "license": "MIT"
     },
+    "node_modules/@standard-schema/spec": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@standard-schema/spec/-/spec-1.1.0.tgz",
+      "integrity": "sha512-l2aFy5jALhniG5HgqrD6jXLi/rUWrKvqN/qJx6yoJsgKhblVd+iqqU4RCXavm/jPityDo5TCvKMnpjKnOriy0w==",
+      "license": "MIT"
+    },
+    "node_modules/@standard-schema/utils": {
+      "version": "0.3.0",
+      "resolved": "https://registry.npmjs.org/@standard-schema/utils/-/utils-0.3.0.tgz",
+      "integrity": "sha512-e7Mew686owMaPJVNNLs55PUvgz371nKgwsc4vxE49zsODpJEnxgxRo2y/OKrqueavXgZNMDVj3DdHFlaSAeU8g==",
+      "license": "MIT"
+    },
     "node_modules/@swc/helpers": {
       "version": "0.5.15",
       "resolved": "https://registry.npmjs.org/@swc/helpers/-/helpers-0.5.15.tgz",
@@ -3804,6 +3843,42 @@
         "react-dom": ">=16.6.0"
       }
     },
+    "node_modules/@tremor/react/node_modules/eventemitter3": {
+      "version": "4.0.7",
+      "resolved": "https://registry.npmjs.org/eventemitter3/-/eventemitter3-4.0.7.tgz",
+      "integrity": "sha512-8guHBZCwKnFhYdHr2ysuRWErTwhoN2X8XELRlrRwpmfeY2jjuUN4taQMsULKUVo1K4DvZl+0pgfyoysHxvmvEw==",
+      "license": "MIT"
+    },
+    "node_modules/@tremor/react/node_modules/react-is": {
+      "version": "18.3.1",
+      "resolved": "https://registry.npmjs.org/react-is/-/react-is-18.3.1.tgz",
+      "integrity": "sha512-/LLMVyas0ljjAtoYiPqYiL8VWXzUUdThrmU5+n20DZv+a+ClRoevUzw5JxU+Ieh5/c87ytoTBV9G1FiKfNJdmg==",
+      "license": "MIT"
+    },
+    "node_modules/@tremor/react/node_modules/recharts": {
+      "version": "2.15.4",
+      "resolved": "https://registry.npmjs.org/recharts/-/recharts-2.15.4.tgz",
+      "integrity": "sha512-UT/q6fwS3c1dHbXv2uFgYJ9BMFHu3fwnd7AYZaEQhXuYQ4hgsxLvsUXzGdKeZrW5xopzDCvuA2N41WJ88I7zIw==",
+      "deprecated": "1.x and 2.x branches are no longer active. Bump to Recharts v3 to receive latest features and bugfixes. See https://github.com/recharts/recharts/wiki/3.0-migration-guide",
+      "license": "MIT",
+      "dependencies": {
+        "clsx": "^2.0.0",
+        "eventemitter3": "^4.0.1",
+        "lodash": "^4.17.21",
+        "react-is": "^18.3.1",
+        "react-smooth": "^4.0.4",
+        "recharts-scale": "^0.4.4",
+        "tiny-invariant": "^1.3.1",
+        "victory-vendor": "^36.6.8"
+      },
+      "engines": {
+        "node": ">=14"
+      },
+      "peerDependencies": {
+        "react": "^16.0.0 || ^17.0.0 || ^18.0.0 || ^19.0.0",
+        "react-dom": "^16.0.0 || ^17.0.0 || ^18.0.0 || ^19.0.0"
+      }
+    },
     "node_modules/@tremor/react/node_modules/tailwind-merge": {
       "version": "2.6.1",
       "resolved": "https://registry.npmjs.org/tailwind-merge/-/tailwind-merge-2.6.1.tgz",
@@ -3814,6 +3889,28 @@
         "url": "https://github.com/sponsors/dcastil"
       }
     },
+    "node_modules/@tremor/react/node_modules/victory-vendor": {
+      "version": "36.9.2",
+      "resolved": "https://registry.npmjs.org/victory-vendor/-/victory-vendor-36.9.2.tgz",
+      "integrity": "sha512-PnpQQMuxlwYdocC8fIJqVXvkeViHYzotI+NJrCuav0ZYFoq912ZHBk3mCeuj+5/VpodOjPe1z0Fk2ihgzlXqjQ==",
+      "license": "MIT AND ISC",
+      "dependencies": {
+        "@types/d3-array": "^3.0.3",
+        "@types/d3-ease": "^3.0.0",
+        "@types/d3-interpolate": "^3.0.1",
+        "@types/d3-scale": "^4.0.2",
+        "@types/d3-shape": "^3.1.0",
+        "@types/d3-time": "^3.0.0",
+        "@types/d3-timer": "^3.0.0",
+        "d3-array": "^3.1.6",
+        "d3-ease": "^3.0.1",
+        "d3-interpolate": "^3.0.1",
+        "d3-scale": "^4.0.2",
+        "d3-shape": "^3.1.0",
+        "d3-time": "^3.0.0",
+        "d3-timer": "^3.0.1"
+      }
+    },
     "node_modules/@tybys/wasm-util": {
       "version": "0.10.3",
       "resolved": "https://registry.npmjs.org/@tybys/wasm-util/-/wasm-util-0.10.3.tgz",
@@ -4069,6 +4166,12 @@
       "integrity": "sha512-ko/gIFJRv177XgZsZcBwnqJN5x/Gien8qNOn0D5bQU/zAzVf9Zt3BlcUiLqhV9y4ARk0GbT3tnUiPNgnTXzc/Q==",
       "license": "MIT"
     },
+    "node_modules/@types/use-sync-external-store": {
+      "version": "0.0.6",
+      "resolved": "https://registry.npmjs.org/@types/use-sync-external-store/-/use-sync-external-store-0.0.6.tgz",
+      "integrity": "sha512-zFDAD+tlpf2r4asuHEj0XH6pY6i0g5NeAHPn+15wk3BV6JA69eERFXC1gyGThDkVa1zCyKr5jox1+2LbV/AMLg==",
+      "license": "MIT"
+    },
     "node_modules/@typescript-eslint/eslint-plugin": {
       "version": "8.60.1",
       "resolved": "https://registry.npmjs.org/@typescript-eslint/eslint-plugin/-/eslint-plugin-8.60.1.tgz",
@@ -6309,6 +6412,16 @@
         "url": "https://github.com/sponsors/ljharb"
       }
     },
+    "node_modules/es-toolkit": {
+      "version": "1.49.0",
+      "resolved": "https://registry.npmjs.org/es-toolkit/-/es-toolkit-1.49.0.tgz",
+      "integrity": "sha512-G5iZ6Pc/FNRY/soKZHC+TxGDD83rHUDXxzaWhGCX44vAv/tMs56WMusnm/KMNK+luUPsgA9U28cGr4RDlSzL2g==",
+      "license": "MIT",
+      "workspaces": [
+        "docs",
+        "benchmarks"
+      ]
+    },
     "node_modules/esbuild": {
       "version": "0.28.1",
       "resolved": "https://registry.npmjs.org/esbuild/-/esbuild-0.28.1.tgz",
@@ -6872,9 +6985,9 @@
       }
     },
     "node_modules/eventemitter3": {
-      "version": "4.0.7",
-      "resolved": "https://registry.npmjs.org/eventemitter3/-/eventemitter3-4.0.7.tgz",
-      "integrity": "sha512-8guHBZCwKnFhYdHr2ysuRWErTwhoN2X8XELRlrRwpmfeY2jjuUN4taQMsULKUVo1K4DvZl+0pgfyoysHxvmvEw==",
+      "version": "5.0.4",
+      "resolved": "https://registry.npmjs.org/eventemitter3/-/eventemitter3-5.0.4.tgz",
+      "integrity": "sha512-mlsTRyGaPBjPedk6Bvw+aqbsXDtoAyAzm5MO7JgU+yVRyMQ5O8bD4Kcci7BS85f93veegeCPkL8R4GLClnjLFw==",
       "license": "MIT"
     },
     "node_modules/expect-type": {
@@ -6901,9 +7014,9 @@
       "license": "MIT"
     },
     "node_modules/fast-equals": {
-      "version": "5.4.0",
-      "resolved": "https://registry.npmjs.org/fast-equals/-/fast-equals-5.4.0.tgz",
-      "integrity": "sha512-jt2DW/aNFNwke7AUd+Z+e6pz39KO5rzdbbFCg2sGafS4mk13MI7Z8O5z9cADNn5lhGODIgLwug6TZO2ctf7kcw==",
+      "version": "5.4.1",
+      "resolved": "https://registry.npmjs.org/fast-equals/-/fast-equals-5.4.1.tgz",
+      "integrity": "sha512-DjlFSM5Pk9cGcL0q5QXl66eGzx0N6szNgaswwc5ZphlBohjTVJSnGgI+rJVOgOi65qUoQnDZN4nDqi33udtydQ==",
       "license": "MIT",
       "engines": {
         "node": ">=6.0.0"
@@ -7688,6 +7801,16 @@
         "node": ">= 4"
       }
     },
+    "node_modules/immer": {
+      "version": "11.1.11",
+      "resolved": "https://registry.npmjs.org/immer/-/immer-11.1.11.tgz",
+      "integrity": "sha512-qzXuyXAkPySAGYkfsAwodDPWT8Zm7/Uo5BNt4BjhMhG5WlWyZZ4wQqnWwdS8kjlQ1Cwu6gjw3A6+0gTQwlyYtw==",
+      "license": "MIT",
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/immer"
+      }
+    },
     "node_modules/import-fresh": {
       "version": "3.3.1",
       "resolved": "https://registry.npmjs.org/import-fresh/-/import-fresh-3.3.1.tgz",
@@ -11589,7 +11712,6 @@
       "version": "17.0.2",
       "resolved": "https://registry.npmjs.org/react-is/-/react-is-17.0.2.tgz",
       "integrity": "sha512-w2GsyukL62IJnlaff/nRegPQR94C/XXamvMWmSHRJ4y7Ts/4ocGRmTHvOs8PSE6pB3dWOrD/nueuU5sduBsQ4w==",
-      "dev": true,
       "license": "MIT"
     },
     "node_modules/react-json-view-lite": {
@@ -11631,6 +11753,29 @@
         "react": ">=18"
       }
     },
+    "node_modules/react-redux": {
+      "version": "9.3.0",
+      "resolved": "https://registry.npmjs.org/react-redux/-/react-redux-9.3.0.tgz",
+      "integrity": "sha512-KQopgqFo/p/fgmAs5qz6p5RWaNAzq40WAu7fJIXnQpYxFPbJYtsJPWvGeF2rOBaY/kEuV77AVsX8TsQzKm+A/g==",
+      "license": "MIT",
+      "dependencies": {
+        "@types/use-sync-external-store": "^0.0.6",
+        "use-sync-external-store": "^1.4.0"
+      },
+      "peerDependencies": {
+        "@types/react": "^18.2.25 || ^19",
+        "react": "^18.0 || ^19",
+        "redux": "^5.0.0"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "redux": {
+          "optional": true
+        }
+      }
+    },
     "node_modules/react-smooth": {
       "version": "4.0.4",
       "resolved": "https://registry.npmjs.org/react-smooth/-/react-smooth-4.0.4.tgz",
@@ -11707,26 +11852,33 @@
       }
     },
     "node_modules/recharts": {
-      "version": "2.15.4",
-      "resolved": "https://registry.npmjs.org/recharts/-/recharts-2.15.4.tgz",
-      "integrity": "sha512-UT/q6fwS3c1dHbXv2uFgYJ9BMFHu3fwnd7AYZaEQhXuYQ4hgsxLvsUXzGdKeZrW5xopzDCvuA2N41WJ88I7zIw==",
+      "version": "3.9.2",
+      "resolved": "https://registry.npmjs.org/recharts/-/recharts-3.9.2.tgz",
+      "integrity": "sha512-G4fy+Pk46RaXgwWMh+Nzhyo/lbFAVqXo9gtetlyehe6Ehge9CsgDuOTwQDD+i1+llaLktNBiNq4bhnGlDRXFtw==",
       "license": "MIT",
+      "workspaces": [
+        "www"
+      ],
       "dependencies": {
-        "clsx": "^2.0.0",
-        "eventemitter3": "^4.0.1",
-        "lodash": "^4.17.21",
-        "react-is": "^18.3.1",
-        "react-smooth": "^4.0.4",
-        "recharts-scale": "^0.4.4",
-        "tiny-invariant": "^1.3.1",
-        "victory-vendor": "^36.6.8"
+        "@reduxjs/toolkit": "^1.9.0 || 2.x.x",
+        "clsx": "^2.1.1",
+        "decimal.js-light": "^2.5.1",
+        "es-toolkit": "^1.39.3",
+        "eventemitter3": "^5.0.1",
+        "immer": "^11.1.8",
+        "react-redux": "8.x.x || 9.x.x",
+        "reselect": "5.2.0",
+        "tiny-invariant": "^1.3.3",
+        "use-sync-external-store": "^1.2.2",
+        "victory-vendor": "^37.0.2"
       },
       "engines": {
-        "node": ">=14"
+        "node": ">=18"
       },
       "peerDependencies": {
-        "react": "^16.0.0 || ^17.0.0 || ^18.0.0 || ^19.0.0",
-        "react-dom": "^16.0.0 || ^17.0.0 || ^18.0.0 || ^19.0.0"
+        "react": "^16.8.0 || ^17.0.0 || ^18.0.0 || ^19.0.0",
+        "react-dom": "^16.0.0 || ^17.0.0 || ^18.0.0 || ^19.0.0",
+        "react-is": "^16.8.0 || ^17.0.0 || ^18.0.0 || ^19.0.0"
       }
     },
     "node_modules/recharts-scale": {
@@ -11738,12 +11890,6 @@
         "decimal.js-light": "^2.4.1"
       }
     },
-    "node_modules/recharts/node_modules/react-is": {
-      "version": "18.3.1",
-      "resolved": "https://registry.npmjs.org/react-is/-/react-is-18.3.1.tgz",
-      "integrity": "sha512-/LLMVyas0ljjAtoYiPqYiL8VWXzUUdThrmU5+n20DZv+a+ClRoevUzw5JxU+Ieh5/c87ytoTBV9G1FiKfNJdmg==",
-      "license": "MIT"
-    },
     "node_modules/redent": {
       "version": "3.0.0",
       "resolved": "https://registry.npmjs.org/redent/-/redent-3.0.0.tgz",
@@ -11758,6 +11904,21 @@
         "node": ">=8"
       }
     },
+    "node_modules/redux": {
+      "version": "5.0.1",
+      "resolved": "https://registry.npmjs.org/redux/-/redux-5.0.1.tgz",
+      "integrity": "sha512-M9/ELqF6fy8FwmkpnF0S3YKOqMyoWJ4+CS5Efg2ct3oY9daQvd/Pc71FpGZsVsbl3Cpb+IIcjBDUnnyBdQbq4w==",
+      "license": "MIT"
+    },
+    "node_modules/redux-thunk": {
+      "version": "3.1.0",
+      "resolved": "https://registry.npmjs.org/redux-thunk/-/redux-thunk-3.1.0.tgz",
+      "integrity": "sha512-NW2r5T6ksUKXCabzhL9z+h206HQw/NJkcLm1GPImRQ8IzfXwRGqjVhKJGauHirT0DAuyy6hjdnMZaRoAcy0Klw==",
+      "license": "MIT",
+      "peerDependencies": {
+        "redux": "^5.0.0"
+      }
+    },
     "node_modules/reflect.getprototypeof": {
       "version": "1.0.10",
       "resolved": "https://registry.npmjs.org/reflect.getprototypeof/-/reflect.getprototypeof-1.0.10.tgz",
@@ -13429,9 +13590,9 @@
       }
     },
     "node_modules/victory-vendor": {
-      "version": "36.9.2",
-      "resolved": "https://registry.npmjs.org/victory-vendor/-/victory-vendor-36.9.2.tgz",
-      "integrity": "sha512-PnpQQMuxlwYdocC8fIJqVXvkeViHYzotI+NJrCuav0ZYFoq912ZHBk3mCeuj+5/VpodOjPe1z0Fk2ihgzlXqjQ==",
+      "version": "37.3.6",
+      "resolved": "https://registry.npmjs.org/victory-vendor/-/victory-vendor-37.3.6.tgz",
+      "integrity": "sha512-SbPDPdDBYp+5MJHhBCAyI7wKM3d5ivekigc2Dk2s7pgbZ9wIgIBYGVw4zGHBml/qTFbexrofXW6Gu4noGxrOwQ==",
       "license": "MIT AND ISC",
       "dependencies": {
         "@types/d3-array": "^3.0.3",

ui/litellm-dashboard/package.json

@@ -50,6 +50,7 @@
     "react-json-view-lite": "2.5.0",
     "react-markdown": "9.1.0",
     "react-syntax-highlighter": "15.6.6",
+    "recharts": "3.9.2",
     "remark-gfm": "4.0.1",
     "tailwind-merge": "3.4.0",
     "uuid": "14.0.0"

ui/litellm-dashboard/src/app/(dashboard)/access-groups/page.tsx

@@ -1,6 +1,6 @@
 "use client";
 
-import { AccessGroupsPage } from "./components/AccessGroupsPage";
+import { AccessGroupsPage } from "./_components/AccessGroupsPage";
 import useAuthorized from "@/app/(dashboard)/hooks/useAuthorized";
 
 export default function AccessGroups() {

ui/litellm-dashboard/src/app/(dashboard)/admin-panel/_components/AdminPanel.test.tsx

@@ -8,34 +8,34 @@ const mockGetAllowedIPs = vi.fn();
 const mockAddAllowedIP = vi.fn();
 const mockDeleteAllowedIP = vi.fn();
 
-vi.mock("./networking", () => ({
+vi.mock("@/components/networking", () => ({
   getSSOSettings: (...args: unknown[]) => mockGetSSOSettings(...args),
   getAllowedIPs: (...args: unknown[]) => mockGetAllowedIPs(...args),
   addAllowedIP: (...args: unknown[]) => mockAddAllowedIP(...args),
   deleteAllowedIP: (...args: unknown[]) => mockDeleteAllowedIP(...args),
 }));
 
-vi.mock("./constants", () => ({
+vi.mock("@/components/constants", () => ({
   useBaseUrl: () => "http://localhost:4000",
 }));
 
-vi.mock("./Settings/AdminSettings/SSOSettings/SSOSettings", () => ({
+vi.mock("@/components/Settings/AdminSettings/SSOSettings/SSOSettings", () => ({
   default: () => <div>SSO Settings</div>,
 }));
 
-vi.mock("./Settings/AdminSettings/UISettings/UISettings", () => ({
+vi.mock("@/components/Settings/AdminSettings/UISettings/UISettings", () => ({
   default: () => <div>UI Settings</div>,
 }));
 
-vi.mock("./SCIM", () => ({
+vi.mock("@/components/SCIM", () => ({
   default: () => <div>SCIM Config</div>,
 }));
 
-vi.mock("./SSOModals", () => ({
+vi.mock("@/components/SSOModals", () => ({
   default: () => <div>SSO Modals</div>,
 }));
 
-vi.mock("./UIAccessControlForm", () => ({
+vi.mock("@/components/UIAccessControlForm", () => ({
   default: () => <div>UI Access Control Form</div>,
 }));
 

ui/litellm-dashboard/src/app/(dashboard)/admin-panel/_components/AdminPanel.tsx

@@ -16,18 +16,18 @@ import {
 } from "@tremor/react";
 import { Alert, Button as Button2, Form, Input, Modal, Space, Tabs, Typography } from "antd";
 import React, { useEffect, useState } from "react";
-import NewBadge from "./common_components/NewBadge";
-import { useBaseUrl } from "./constants";
-import NotificationsManager from "./molecules/notifications_manager";
-import { addAllowedIP, deleteAllowedIP, getAllowedIPs, getSSOSettings } from "./networking";
-import SCIMConfig from "./SCIM";
-import LoggingSettings from "./Settings/AdminSettings/LoggingSettings/LoggingSettings";
-import SSOSettings from "./Settings/AdminSettings/SSOSettings/SSOSettings";
-import UISettings from "./Settings/AdminSettings/UISettings/UISettings";
-import HashicorpVault from "./Settings/AdminSettings/HashicorpVault/HashicorpVault";
-import PluginSettings from "./Settings/AdminSettings/PluginSettings/PluginSettings";
-import SSOModals from "./SSOModals";
-import UIAccessControlForm from "./UIAccessControlForm";
+import NewBadge from "@/components/common_components/NewBadge";
+import { useBaseUrl } from "@/components/constants";
+import NotificationsManager from "@/components/molecules/notifications_manager";
+import { addAllowedIP, deleteAllowedIP, getAllowedIPs, getSSOSettings } from "@/components/networking";
+import SCIMConfig from "@/components/SCIM";
+import LoggingSettings from "@/components/Settings/AdminSettings/LoggingSettings/LoggingSettings";
+import SSOSettings from "@/components/Settings/AdminSettings/SSOSettings/SSOSettings";
+import UISettings from "@/components/Settings/AdminSettings/UISettings/UISettings";
+import HashicorpVault from "@/components/Settings/AdminSettings/HashicorpVault/HashicorpVault";
+import PluginSettings from "@/components/Settings/AdminSettings/PluginSettings/PluginSettings";
+import SSOModals from "@/components/SSOModals";
+import UIAccessControlForm from "@/components/UIAccessControlForm";
 
 const { Title, Paragraph, Text } = Typography;
 

ui/litellm-dashboard/src/app/(dashboard)/admin-panel/page.tsx

@@ -1,6 +1,6 @@
 "use client";
 
-import AdminPanel from "@/components/AdminPanel";
+import AdminPanel from "./_components/AdminPanel";
 import useAuthorized from "@/app/(dashboard)/hooks/useAuthorized";
 import useProxySettings from "@/app/(dashboard)/hooks/proxySettings/useProxySettings";
 

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/add_agent_form.tsx

@@ -4,7 +4,7 @@ import MessageManager from "@/components/molecules/message_manager";
 import { resolveLogoSrc } from "@/lib/assetPaths";
 import { Button } from "@tremor/react";
 import { CheckCircleFilled, KeyOutlined, RobotOutlined, AppstoreOutlined, InfoCircleOutlined } from "@ant-design/icons";
-import CreatedKeyDisplay from "../shared/CreatedKeyDisplay";
+import CreatedKeyDisplay from "@/components/shared/CreatedKeyDisplay";
 import {
   createAgentCall,
   getAgentCreateMetadata,
@@ -14,19 +14,19 @@ import {
   keyUpdateCall,
   modelAvailableCall,
   AgentCreateInfo,
-} from "../networking";
+} from "@/components/networking";
 import useAuthorized from "@/app/(dashboard)/hooks/useAuthorized";
-import { getModelDisplayName } from "../key_team_helpers/fetch_available_models_team_key";
-import { Team } from "../key_team_helpers/key_list";
-import TeamDropdown from "../common_components/team_dropdown";
+import { getModelDisplayName } from "@/components/key_team_helpers/fetch_available_models_team_key";
+import { Team } from "@/components/key_team_helpers/key_list";
+import TeamDropdown from "@/components/common_components/team_dropdown";
 import AgentFormFields from "./agent_form_fields";
 import AgentCardDiscovery, { DiscoveredAgentCardSelection } from "./agent_card_discovery";
 import { buildDiscoveryRequest, overlayDiscoveredCardParams } from "./agent_discovery_utils";
 import DynamicAgentFormFields, { buildDynamicAgentData } from "./dynamic_agent_form_fields";
 import { getDefaultFormValues, buildAgentDataFromForm } from "./agent_config";
-import MCPServerSelector from "../mcp_server_management/MCPServerSelector";
-import MCPToolPermissions from "../mcp_server_management/MCPToolPermissions";
-import GuardrailSelector from "../guardrails/GuardrailSelector";
+import MCPServerSelector from "@/components/mcp_server_management/MCPServerSelector";
+import MCPToolPermissions from "@/components/mcp_server_management/MCPToolPermissions";
+import GuardrailSelector from "@/components/guardrails/GuardrailSelector";
 
 const { Step } = Steps;
 

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_card_discovery.test.tsx

@@ -2,18 +2,18 @@ import React from "react";
 import { describe, it, expect, vi, beforeEach, afterEach } from "vitest";
 import { screen, waitFor } from "@testing-library/react";
 import userEvent from "@testing-library/user-event";
-import { renderWithProviders } from "../../../tests/test-utils";
+import { renderWithProviders } from "@/../tests/test-utils";
 import AgentCardDiscovery from "./agent_card_discovery";
 
-vi.mock("../networking", async () => {
-  const actual = await vi.importActual<any>("../networking");
+vi.mock("@/components/networking", async () => {
+  const actual = await vi.importActual<any>("@/components/networking");
   return {
     ...actual,
     discoverAgentCardCall: vi.fn(),
   };
 });
 
-import { discoverAgentCardCall } from "../networking";
+import { discoverAgentCardCall } from "@/components/networking";
 
 const mockDiscover = discoverAgentCardCall as unknown as ReturnType<typeof vi.fn>;
 

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_card_discovery.tsx

@@ -11,7 +11,7 @@ import {
   SearchOutlined,
 } from "@ant-design/icons";
 
-import { DiscoveredAgentCard, discoverAgentCardCall } from "../networking";
+import { DiscoveredAgentCard, discoverAgentCardCall } from "@/components/networking";
 import {
   ALLOWED_CAPABILITY_KEYS,
   selectionsFromSavedAgentCard,

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_cost_view.tsx

@@ -1,7 +1,7 @@
 import React from "react";
 import { Title } from "@tremor/react";
 import { Descriptions } from "antd";
-import { Agent } from "./types";
+import { Agent } from "@/components/agents/types";
 
 interface AgentCostViewProps {
   agent: Agent;

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_discovery_utils.ts

@@ -1,4 +1,4 @@
-import { AgentCreateInfo, DiscoveredAgentCard, DiscoveryMode } from "../networking";
+import { AgentCreateInfo, DiscoveredAgentCard, DiscoveryMode } from "@/components/networking";
 
 export interface DiscoveryRequestPlan {
   url: string;

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_info.tsx

@@ -3,11 +3,11 @@ import { Card, Title, Text, Button as TremorButton, Tab, TabGroup, TabList, TabP
 import { Form, Input, InputNumber, Button as AntButton, Spin, Descriptions, Divider } from "antd";
 import MessageManager from "@/components/molecules/message_manager";
 import { ArrowLeftIcon } from "@heroicons/react/outline";
-import { getAgentInfo, patchAgentCall, getAgentCreateMetadata, AgentCreateInfo } from "../networking";
-import { Agent } from "./types";
-import { KeyResponse } from "../key_team_helpers/key_list";
+import { getAgentInfo, patchAgentCall, getAgentCreateMetadata, AgentCreateInfo } from "@/components/networking";
+import { Agent } from "@/components/agents/types";
+import { KeyResponse } from "@/components/key_team_helpers/key_list";
 import { useKeys } from "@/app/(dashboard)/hooks/keys/useKeys";
-import KeyInfoView from "../templates/key_info_view";
+import KeyInfoView from "@/components/templates/key_info_view";
 import AgentVirtualKeys from "./agent_virtual_keys";
 import AgentFormFields from "./agent_form_fields";
 import DynamicAgentFormFields, { buildDynamicAgentData } from "./dynamic_agent_form_fields";

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_type_utils.ts

@@ -1,5 +1,5 @@
-import { Agent } from "./types";
-import { AgentCreateInfo } from "../networking";
+import { Agent } from "@/components/agents/types";
+import { AgentCreateInfo } from "@/components/networking";
 
 /**
  * Detects the agent type from an agent's litellm_params.

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_virtual_keys.test.tsx

@@ -2,9 +2,9 @@ import React from "react";
 import { describe, it, expect, vi } from "vitest";
 import { screen } from "@testing-library/react";
 import userEvent from "@testing-library/user-event";
-import { renderWithProviders } from "../../../tests/test-utils";
+import { renderWithProviders } from "@/../tests/test-utils";
 import AgentVirtualKeys from "./agent_virtual_keys";
-import type { KeyResponse } from "../key_team_helpers/key_list";
+import type { KeyResponse } from "@/components/key_team_helpers/key_list";
 
 const makeKey = (overrides: Partial<KeyResponse>): KeyResponse =>
   ({

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_virtual_keys.tsx

@@ -1,7 +1,7 @@
 import React from "react";
 import { Button, Tooltip, Typography } from "antd";
 import { KeyOutlined } from "@ant-design/icons";
-import { KeyResponse } from "../key_team_helpers/key_list";
+import { KeyResponse } from "@/components/key_team_helpers/key_list";
 
 const { Title, Text } = Typography;
 

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/dynamic_agent_form_fields.tsx

@@ -1,6 +1,6 @@
 import React from "react";
 import { Form, Input, Select, Collapse } from "antd";
-import { AgentCreateInfo, AgentCredentialFieldMetadata } from "../networking";
+import { AgentCreateInfo, AgentCredentialFieldMetadata } from "@/components/networking";
 import { AGENT_FORM_CONFIG } from "./agent_config";
 import CostConfigFields from "./cost_config_fields";
 

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/index.test.tsx

@@ -1,19 +1,19 @@
 import React from "react";
 import { render, screen, waitFor, act, fireEvent, within } from "@testing-library/react";
 import { describe, it, expect, vi, beforeEach } from "vitest";
-import AgentsPanel from "./agents";
-import * as networking from "./networking";
+import AgentsPanel from "./index";
+import * as networking from "@/components/networking";
 
-vi.mock("./networking", () => ({
+vi.mock("@/components/networking", () => ({
   getAgentsList: vi.fn().mockResolvedValue({ agents: [] }),
   deleteAgentCall: vi.fn(),
 }));
 
-vi.mock("./agents/add_agent_form", () => ({
+vi.mock("./add_agent_form", () => ({
   default: () => <div data-testid="add-agent-form" />,
 }));
 
-vi.mock("./agents/agent_info", () => ({
+vi.mock("./agent_info", () => ({
   default: () => <div data-testid="agent-info" />,
 }));
 

ui/litellm-dashboard/src/app/(dashboard)/agents/_components/index.tsx

@@ -13,15 +13,15 @@ import {
 } from "@tremor/react";
 import { Modal, Alert, Tooltip, Skeleton, Switch } from "antd";
 import { CheckCircleOutlined } from "@ant-design/icons";
-import { getAgentsList, deleteAgentCall } from "./networking";
-import AddAgentForm from "./agents/add_agent_form";
+import { getAgentsList, deleteAgentCall } from "@/components/networking";
+import AddAgentForm from "./add_agent_form";
 import { isAdminRole } from "@/utils/roles";
-import AgentInfoView from "./agents/agent_info";
-import NotificationsManager from "./molecules/notifications_manager";
-import { Agent } from "./agents/types";
-import { Team } from "./key_team_helpers/key_list";
+import AgentInfoView from "./agent_info";
+import NotificationsManager from "@/components/molecules/notifications_manager";
+import { Agent } from "@/components/agents/types";
+import { Team } from "@/components/key_team_helpers/key_list";
 import { DateCell, IdCell, MoneyCell, StatusBadge } from "@/components/shared/table_cells";
-import TableIconActionButton from "./common_components/IconActionButton/TableIconActionButtons/TableIconActionButton";
+import TableIconActionButton from "@/components/common_components/IconActionButton/TableIconActionButtons/TableIconActionButton";
 
 interface AgentsPanelProps {
   accessToken: string | null;