fix(proxy): restore Claude gateway model discovery (#48204) (#48221)

* fix(proxy): restore Claude gateway model discovery (#48204)

* test(proxy): cover gateway discovery after OAuth sign-in (#48204)

* refactor(proxy): share model listing route policy (#48204)

* test(proxy): isolate SSO settings environment (#48204)

enterprise/litellm_enterprise/proxy/auth/route_checks.py

@@ -2,6 +2,8 @@ import os
 
 from fastapi import HTTPException, status
 
+from litellm.proxy._types import MODEL_LIST_ROUTES
+
 
 class EnterpriseRouteChecks:
     @staticmethod
@@ -43,7 +45,7 @@ class EnterpriseRouteChecks:
 
     # Routes that should remain accessible even when LLM API endpoints are disabled.
     # These are read-only model listing routes needed by the Admin UI.
-    LLM_API_EXEMPT_ROUTES = ["/models", "/v1/models"]
+    LLM_API_EXEMPT_ROUTES = MODEL_LIST_ROUTES
 
     @staticmethod
     def should_call_route(route: str):

litellm/proxy/_types.py

@@ -219,6 +219,8 @@ from litellm.types.proxy.auth.user_roles import (  # noqa: E402  # re-export aft
     KeyManagementRoutes as KeyManagementRoutes,  # noqa: PLC0414  # public re-export
 )
 
+MODEL_LIST_ROUTES: Final = ("/models", "/v1/models", "/claude_code_gateway/v1/models")
+
 
 class LiteLLMRoutes(enum.Enum):
     openai_route_names = [
@@ -307,8 +309,7 @@ class LiteLLMRoutes(enum.Enum):
         "/threads/{thread_id}/runs",
         "/v1/threads/{thread_id}/runs",
         # models
-        "/models",
-        "/v1/models",
+        *MODEL_LIST_ROUTES,
         # token counter
         "/utils/token_counter",
         "/utils/model_info",
@@ -577,8 +578,7 @@ class LiteLLMRoutes(enum.Enum):
         "/health/services",
         "/key/list",
         "/user/filter/ui",
-        "/models",
-        "/v1/models",
+        *MODEL_LIST_ROUTES,
         "/sso/get/ui_settings",
         "/get/user_banner",
         "/get/latest_release_info",
@@ -760,8 +760,7 @@ class LiteLLMRoutes(enum.Enum):
         "/model/info",
         "/v2/model/info",
         "/v2/key/info",
-        "/models",
-        "/v1/models",
+        *MODEL_LIST_ROUTES,
         "/global/spend",
         "/global/spend/logs",
         "/global/spend/keys",

litellm/proxy/anthropic_endpoints/gateway_endpoints.py

@@ -58,6 +58,7 @@ from litellm.proxy.common_utils.http_parsing_utils import (  # noqa: F401  # leg
 )
 from litellm.proxy.management_endpoints.sso_helper_utils import CLI_SSO_SESSIONS_TARGET
 from litellm.proxy.management_endpoints.ui_sso import CliSsoTeamDetail
+from litellm.proxy.proxy_server import model_list
 from litellm.types.llms.base import LiteLLMBaseModel
 
 GATEWAY_PREFIX: Final = "/claude_code_gateway"
@@ -158,6 +159,13 @@ router: Final = APIRouter(
 _GATEWAY_ENABLED: Final = (Depends(ensure_gateway_enabled),)
 _AUTHENTICATED: Final = (Depends(user_api_key_auth),)
 
+router.add_api_route(
+    "/v1/models",
+    model_list,
+    methods=["GET"],
+    dependencies=_GATEWAY_ENABLED,
+    include_in_schema=False,
+)
 router.add_api_route(
     "/v1/messages",
     anthropic_response,

litellm/proxy/auth/auth_checks.py

@@ -47,6 +47,7 @@ from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
 from litellm.litellm_core_utils.safe_json_loads import safe_json_loads
 from litellm.models.project import LiteLLM_ProjectTable
 from litellm.proxy._types import (
+    MODEL_LIST_ROUTES,
     RBAC_ROLES,
     UI_TEAM_ID,
     CallInfo,
@@ -910,8 +911,7 @@ def _global_proxy_budget_check(global_proxy_spend: float | None, skip_budget_che
         and not skip_budget_checks
         and global_proxy_spend is not None
         and RouteChecks.is_llm_api_route(route=route)
-        and route != "/v1/models"
-        and route != "/models"
+        and route not in MODEL_LIST_ROUTES
     ):
         if math.isfinite(litellm.max_budget) and global_proxy_spend > litellm.max_budget:
             raise litellm.BudgetExceededError(
@@ -1028,8 +1028,7 @@ async def check_tools_allowlist(
 # budget cannot reach side-effectful routes like /health/services (Slack/email/webhook). See #27923.
 MODEL_DISCOVERY_ROUTES: Final = frozenset(
     {
-        "/v1/models",
-        "/models",
+        *MODEL_LIST_ROUTES,
         "/model/info",
         "/v1/model/info",
         "/v2/model/info",

litellm/proxy/auth/route_checks.py

@@ -7,6 +7,7 @@ from fastapi import HTTPException, Request, status
 
 from litellm._logging import verbose_proxy_logger
 from litellm.proxy._types import (
+    MODEL_LIST_ROUTES,
     UI_TEAM_ID,
     CommonProxyErrors,
     KeyManagementRoutes,
@@ -93,8 +94,7 @@ _SELF_SCOPED_CREDENTIAL_ROUTES: Final = (
 
 _UNBILLED_ROUTES: Final[frozenset[str]] = frozenset(
     {
-        "/models",
-        "/v1/models",
+        *MODEL_LIST_ROUTES,
         "/utils/token_counter",
         "/utils/transform_request",
         "/responses/input_tokens",

tests/integration/authorization/test_claude_code_gateway_oauth.py

@@ -15,6 +15,7 @@ from urllib.parse import parse_qs, urlencode, urlsplit
 import httpx
 import pytest
 import yaml
+from anthropic import Anthropic
 from anthropic.types import Message
 from integration._support import claude_code as cc
 from integration._support.client import Gateway, eventually, gateway_from_environment
@@ -397,7 +398,31 @@ def test_device_authorization_polls_complete_sso_and_mints_a_model_restricted_be
                 "anthropic-beta": cc.CLI_BETA,
                 "x-claude-code-session-id": f"oauth-{user_id}",
             }
-            request_body: Final = {**request_template, "model": model_a}
+            standard_listing: Final = enabled_gateway.client.get("/v1/models", headers=bearer_headers)
+            gateway_listing: Final = enabled_gateway.client.get(
+                "/claude_code_gateway/v1/models", headers=bearer_headers
+            )
+            assert standard_listing.status_code == 200, standard_listing.text
+            assert gateway_listing.status_code == 200, gateway_listing.text
+            assert gateway_listing.json() == standard_listing.json()
+            with Anthropic(
+                base_url=f"{str(enabled_gateway.client.base_url).rstrip('/')}/claude_code_gateway",
+                auth_token=access_token.access_token,
+            ) as sdk:
+                discovered_models: Final = tuple(sdk.models.list())
+            assert {model.id for model in discovered_models} == {model_a, model_c}
+            discovered_model: Final = next(model for model in discovered_models if model.id == model_a)
+            openai_listing: Final = enabled_gateway.client.get(
+                "/claude_code_gateway/v1/models",
+                headers={"authorization": f"Bearer {access_token.access_token}"},
+            )
+            assert openai_listing.status_code == 200, openai_listing.text
+            assert openai_listing.json()["object"] == "list"
+            assert {model["id"] for model in openai_listing.json()["data"]} == {model_a, model_c}
+            unauthenticated: Final = enabled_gateway.client.get("/claude_code_gateway/v1/models")
+            assert unauthenticated.status_code == 401, unauthenticated.text
+            assert upstream.drain() == ()
+            request_body: Final = {**request_template, "model": discovered_model.id}
             expected_message: Final = _MESSAGE.validate_python(
                 {
                     **_JSON_OBJECT.validate_json(

tests/unit/proxy/agent_endpoints/auth/test_managed_authorization.py

@@ -588,6 +588,6 @@ def test_registered_inference_routes_have_an_explicit_managed_access_decision(ro
         "/vector_stores", "/vector_store/", "/search", "/containers", "/skills", "/claude-code/",
         "/interactions", "/agents", "/responses/{", "/responses/input_tokens",
         "/realtime/client_secrets", "/realtime/calls", "/realtime/transcription_sessions",
-    )) or normalized in ("/models", "/cursor/models", "/cursor/v1/models")
+    )) or normalized in ("/models", "/cursor/models", "/cursor/v1/models", "/claude_code_gateway/v1/models")
     concrete: Final = route.split("?")[0].replace("{model}", "model").replace("{model_name:path}", "model")
     assert managed_agent_route_allowed(concrete, None) is not unsupported, route

tests/unit/proxy/anthropic_endpoints/test_gateway_endpoints.py

@@ -813,3 +813,95 @@ def test_messages_gated_by_enable_flag():
     with _gateway_env(enabled=False) as (client, _):
         resp = client.post("/claude_code_gateway/v1/messages", json={"model": "claude-sonnet-4-5", "messages": []})
     assert resp.status_code == 404
+
+
+@pytest.mark.parametrize("headers", [{}, {"anthropic-version": "2023-06-01"}])
+def test_gateway_model_discovery_preserves_grants_and_format(
+    monkeypatch: pytest.MonkeyPatch, headers: dict[str, str]
+) -> None:
+    from litellm import Router
+    from litellm.proxy import proxy_server
+    from litellm.proxy._types import LitellmUserRoles, UserAPIKeyAuth
+
+    model_router: Final = Router(
+        model_list=[
+            {
+                "model_name": name,
+                "litellm_params": {"model": f"openai/{name}", "api_key": "local-placeholder"},
+            }
+            for name in ("discovery-example", "restricted-example")
+        ]
+    )
+    monkeypatch.setattr(proxy_server, "llm_router", model_router)
+    monkeypatch.setattr(proxy_server, "llm_model_list", model_router.model_list)
+    monkeypatch.setattr(proxy_server, "prisma_client", None)
+    monkeypatch.setattr(proxy_server, "user_model", None)
+    monkeypatch.setattr(proxy_server, "general_settings", {"enable_claude_code_gateway": True})
+    caller: Final = UserAPIKeyAuth(
+        user_role=LitellmUserRoles.INTERNAL_USER,
+        models=["discovery-example"],
+    )
+    app: Final = FastAPI()
+    app.include_router(proxy_server.router)
+    app.include_router(gateway_endpoints.router)
+    app.dependency_overrides[gateway_endpoints.user_api_key_auth] = lambda: caller
+    with TestClient(app) as client:
+        standard: Final = client.get("/v1/models", headers=headers)
+        gateway: Final = client.get("/claude_code_gateway/v1/models", headers=headers)
+        assert standard.status_code == 200, standard.text
+        assert gateway.status_code == 200, gateway.text
+        assert gateway.json() == standard.json()
+        assert [model["id"] for model in gateway.json()["data"]] == ["discovery-example"]
+        invalid_scope: Final = client.get("/claude_code_gateway/v1/models?scope=invalid", headers=headers)
+        assert invalid_scope.status_code == 400
+
+
+def test_gateway_model_discovery_requires_authentication() -> None:
+    with _gateway_env(real_auth=True) as (client, _), pytest.raises(ProxyException) as exc_info:
+        client.get("/claude_code_gateway/v1/models")
+    assert exc_info.value.code == "401"
+
+
+@pytest.mark.parametrize("settings", [{}, {"enable_claude_code_gateway": False}])
+def test_gateway_model_discovery_disabled(monkeypatch: pytest.MonkeyPatch, settings: dict[str, bool]) -> None:
+    from litellm.proxy import proxy_server
+
+    with _gateway_env(real_auth=True) as (client, _):
+        monkeypatch.setattr(proxy_server, "general_settings", settings)
+        response: Final = client.get(
+            "/claude_code_gateway/v1/models", headers={"Authorization": f"Bearer {_MASTER_KEY}"}
+        )
+        assert response.status_code == 404, response.text
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("route", ["/models", "/v1/models", "/claude_code_gateway/v1/models"])
+async def test_gateway_model_discovery_preserves_route_exemptions(monkeypatch: pytest.MonkeyPatch, route: str) -> None:
+    from litellm_enterprise.proxy.auth.route_checks import EnterpriseRouteChecks
+
+    import litellm
+    from litellm.proxy._types import LiteLLM_TeamTable, UserAPIKeyAuth
+    from litellm.proxy.auth.auth_checks import common_checks
+    from litellm.proxy.auth.route_checks import RouteChecks
+
+    monkeypatch.setattr(litellm, "max_budget", 1.0)
+    result: Final = await common_checks(
+        request_body={},
+        team_object=LiteLLM_TeamTable(team_id="test-team", spend=150.0, max_budget=100.0),
+        user_object=None,
+        end_user_object=None,
+        global_proxy_spend=2.0,
+        general_settings={},
+        route=route,
+        llm_router=None,
+        proxy_logging_obj=MagicMock(),
+        valid_token=UserAPIKeyAuth(team_id="test-team"),
+        request=MagicMock(),
+    )
+    assert result is True
+    assert RouteChecks.is_unbilled_route(route)
+    assert RouteChecks.is_llm_api_route(route)
+    monkeypatch.setenv("DISABLE_LLM_API_ENDPOINTS", "true")
+    monkeypatch.setenv("DISABLE_ADMIN_ENDPOINTS", "true")
+    monkeypatch.setattr("litellm.proxy.proxy_server.premium_user", True)
+    EnterpriseRouteChecks.should_call_route(route)

tests/unit/proxy/ui_crud_endpoints/test_proxy_setting_endpoints.py

@@ -1,6 +1,8 @@
 import json
 import os
+from collections.abc import Iterator
 from typing import Final
+from unittest.mock import patch
 
 import pytest
 from fastapi.testclient import TestClient
@@ -181,6 +183,12 @@ def mock_auth():
     app.dependency_overrides.pop(user_api_key_auth, None)
 
 
+@pytest.fixture(autouse=True)
+def isolated_environment() -> Iterator[None]:
+    with patch.dict(os.environ):
+        yield
+
+
 @pytest.fixture(autouse=True)
 def fresh_settings_store(monkeypatch: pytest.MonkeyPatch) -> None:
     from litellm.proxy import proxy_server