litellm/tests/mcp_tests/test_mcp_server.py
Sameer Kankute 7eacdd5258
chore: litellm oss staging 250626 (#31305)
* fix(anthropic): support Bearer auth for custom api_base endpoints (Fixes #30926)

* style: format common_utils.py with black

* fix(anthropic): extract api_base from litellm_params in batches/files validate_environment

* fix(anthropic): scope Bearer key check to custom api_base endpoints

* fix(streaming): reset Anthropic message_start cursor (output_tokens=1) when no message_delta arrives

The Anthropic streaming protocol emits `message_start.usage.output_tokens=1`
as a placeholder cursor; the real cumulative output count only arrives in
the final `message_delta` event. When a stream is cancelled before
`message_delta` lands (common for thinking models on long-tail prompts),
ChunkProcessor._calculate_usage_per_chunk's last-wins accumulator left
completion_tokens stuck at 1. Because 1 is truthy, the
`completion_tokens or token_counter(text=...)` fallback in
calculate_usage() never fired, and requests were billed for 1 output
token even when several thousand tokens of text had actually streamed.

Fix: track whether any chunk's completion_tokens exceeded 1
(saw_non_cursor_completion). If the only update we saw was the cursor,
reset completion_tokens to 0 so the text-based fallback estimates from
the real completion content.

Legitimate 1-token completions (model returns "Yes." etc.) are unaffected
in practice — token_counter on a 1-token completion_output also yields
~1, so billing stays approximately correct.

Tests:
- TestAnthropicCursorBug (6 cases) — pins the post-fix behavior
- TestNonAnthropicStreamingIntact (2 cases) — guards against regression on
  providers without the cursor pattern

All 8 new tests pass; 9 existing streaming_chunk_builder_utils tests
still pass.

* fix(streaming): scope cursor reset to anthropic provider + recognize message_delta arrival

Addresses both Greptile P2 threads on PR #30420:

CLASS A — Anthropic-specific heuristic was applied globally
============================================================
The `completion_tokens == 1 and not saw_non_cursor_completion` reset
lived in provider-neutral `streaming_chunk_builder_utils.py`. Any
non-Anthropic provider that legitimately reports completion_tokens=1
in a single usage chunk (perfectly normal for short OpenAI / Bedrock /
Vertex single-token replies with stream_options.include_usage=true)
would have its value silently rewritten to 0 and re-billed via
token_counter — producing a different number than what the provider
actually charged.

Fix: gate the reset on `custom_llm_provider == "anthropic"`, resolved
from the first chunk's `_hidden_params` (the same field set by
streaming_handler.py:722 on the live path). Unknown / missing provider
is treated as non-Anthropic and skips the reset, so newer providers and
custom plugins are also safe by default.

CLASS B — `saw_non_cursor_completion` missed legitimate single-token replies
============================================================
Previous condition was `usage_chunk_dict["completion_tokens"] > 1`,
which never fires for an Anthropic stream where the model legitimately
emits exactly one output token (e.g., "Yes."). Anthropic still sends
message_start (output_tokens=1, the cursor) AND message_delta
(output_tokens=1, the real value) — same value, but two distinct usage
events. The old check couldn't tell that apart from a cancelled stream
where only message_start landed.

Fix: track `completion_usage_updates` and flip `saw_non_cursor_completion`
when EITHER (1) the value exceeds 1 (definitely not a placeholder), OR
(2) we've seen >=2 completion-bearing usage events (positive evidence
that message_delta arrived). Cancelled cursor-only streams still have
exactly one event and still hit the reset; cache chunks with
completion_tokens=0 don't count toward the threshold.

Tests
============================================================
- _make_chunk now sets `_hidden_params["custom_llm_provider"]` (default
  "anthropic") so the gate is exercised by every existing test —
  none of them needed assertion changes besides the legitimate-single-
  token case, which now expects exactly 1 (was a fuzzy 0..3 range).
- New: test_anthropic_cache_only_chunks_after_message_start_still_resets
- New: test_non_anthropic_provider_completion_tokens_one_not_reset
- New: test_unknown_provider_completion_tokens_one_not_reset

11/11 tests pass.

* chore: add Co-authored-by trailer for attribution

Co-authored-by: songkuan-zheng <songkuan-zheng@users.noreply.github.com>

* fix(anthropic): preserve messages cache usage

* style(anthropic): format messages cache usage helper

* fix(anthropic): accept integral float cache token counts

Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

* fix(anthropic): accept integral float cache token counts

* test(anthropic): cover cache usage edge cases

* fix(gemini): preserve thoughtSignature for server-side tool responses

When Gemini API returns toolCall and toolResponse parts, they might have
different thoughtSignatures. Previously, LiteLLM merged them into a single
dict, overwriting the response's thoughtSignature with the call's.
This fix extracts them separately and re-injects them correctly.

TAG=agy
CONV=755b21d0-3200-40bc-bd1a-bb58a378a9a6

* fix(gemini): address PR comments on thoughtSignature handling

- Fix orphan-response thoughtSignature regression by copying thought_signature to response_thought_signature
- Add missing assertions in existing tests
- Add new unit tests for orphan-response signature handling

TAG=agy
CONV=755b21d0-3200-40bc-bd1a-bb58a378a9a6

* feat(mcp): include server alias and server_id in mcp_info response

- Add alias and server_id fields to mcp_info object in /mcp-rest/tools/list endpoint
- Update rest_endpoints.py to surface alias from server config
- Add test coverage in test_mcp_server.py and test_rest_endpoints.py

Fixes #31015

* fix(proxy): reject non-finite spend via validate_finite_spend

A NaN/-inf spend would bypass spend >= max_budget enforcement. Add a
shared finite-value guard, defined above the litellm.proxy.* imports to
avoid the module-level cyclic-import warning.

* fix(proxy): require admin for any /key/update spend, reject non-finite

Gate the admin check on the presence of `spend` (not a value diff): the
DB spend lags the live cross-pod counter, so an "unchanged" spend on the
non-admin path let a key owner / team member overwrite the live counter
below real usage. Also reject NaN/+-inf spend before the DB write.

* fix(proxy): invalidate spend counter on /user/update spend change

A direct spend change on /user/update wrote the DB row but left the warm
cross-pod counter at the stale value, so enforcement kept reading the old
spend. Invalidate spend:user:{user_id} after the write (reseed-from-DB),
and reject non-finite spend before the write.

* fix(cache): route Bedrock semantic-cache sync embedding through the Router (#28244)

The semantic cache's embedding model is a proxy Router alias whose AWS
credentials (aws_role_name, aws_session_name) live only in the Router
deployment's litellm_params. The sync embedding paths called litellm.embedding()
directly, bypassing the Router, so they could neither resolve the alias nor
assume the configured role; cross-account Bedrock semantic caching failed with
"bedrock:InvokeModel is not authorized". On Redis this surfaced at proxy startup
because redisvl's CustomTextVectorizer eagerly fires a dimension-probe embedding
during cache construction, while llm_router is still None.

Fix A: make the sync paths mirror the already-correct async paths. A shared,
dependency-injected helper (litellm/caching/_embedding_router.py) decides whether
to route through llm_router.embedding(...) when the model is a Router deployment,
else fall back to direct litellm.embedding(...). Redis and qdrant sync
set_cache/get_cache now precompute the embedding and pass vector= to the backend,
exactly as the async astore/acheck already do. Both async _get_async_embedding
methods are unified onto the same helper and now forward the caller's full
metadata instead of a hand-picked subset.

Fix B (Redis only): defer redisvl index construction from __init__ into a lazy,
memoized llmcache property, so the dimension-probe embedding fires on first cache
use, after llm_router is wired. A failed build is not memoized, so a transient
outage recovers on the next request.

Known limitation: resolve_embedding_router gates on an exact model-name match
(same as the shipped async path); wildcard/alias/team-public routes still fall
back to direct embedding. Tracked as a follow-up.

* fix(cache): harden embedding-router and shrink Any surface (review)

Address review feedback on the semantic-cache aws-role fix (#28244):

- resolve_embedding_router now skips deployment entries missing model_name
  instead of raising KeyError on a malformed model_list (Greptile P2);
  add a regression test that fails on the old direct-key access.
- Replace the `**kwargs: Any` passthrough on the four cache _get_embedding /
  _get_async_embedding helpers with an explicit, typed
  `metadata: Optional[Dict[str, Any]] = None` parameter. The helpers only
  ever consumed kwargs["metadata"], so this is behavior-preserving, makes the
  forwarded field obvious at the call site, and removes three bare-Any
  annotations (keeps the strict-rule ANN401 budget within ceiling).
- Note in _build_llmcache that redisvl's dimension-probe embedding adds one
  extra billable embedding on the first cache request (Greptile P2).

* fix(bedrock_mantle): correct responses routing for openai.gpt-5.x models

Dashboard Test Connection for bedrock_mantle/openai.gpt-5.4 and openai.gpt-5.5 was failing with maximum recursion depth errors and "model does not exist"

Route detection in the bedrock provider matched route tokens by plain substring, so the bedrock_mantle/ prefix was mistaken for the mantle/ invoke route and the body model was rewritten to bedrock_openai.gpt-5.5; route tokens now only match at a path-segment boundary so the bare model name is preserved

A responses-mode model whose provider has no responses config bounced forever between the responses API and chat completions; the responses to completion fallback now tags its call so completion() does not bridge back, breaking the loop

The Test Connection endpoint hardcoded the test mode to chat, which disabled mode auto-detection for responses-only models; the default is now None so the mode is detected from model capabilities

acompletion() now drops a duplicate acompletion kwarg before building the partial and treats model_info=None as an empty dict to avoid a NoneType crash

* test(bedrock_mantle): cover route guard and bridge flag; fix reportArgumentType regression

Adds the regression coverage codecov flagged on the two responses to completion
bridge guard lines and the bedrock route-prefix helper. The handler tests drive
both the sync and async fallback paths with litellm.completion and
litellm.acompletion mocked, and assert the forwarded kwargs carry
_skip_responses_api_bridge=True, so dropping either flag line fails the suite.
The common_utils tests assert that bedrock_mantle/openai.gpt-5.x no longer
resolves to the mantle route while the genuine mantle/ and bedrock/mantle/ ids
still do, exercising both branches of _model_has_route_prefix.

Also aligns update_messages_with_model_file_ids model_id to Optional[str],
matching its Responses API sibling, so the defensive model_info fallback no
longer introduces a new reportArgumentType in completion(); the file-id lookup
narrows model_id before the dict get

* chore(ui): sync generated OpenAPI types for optional test_connection mode

The test_model_connection mode body param default changed from chat to None so
the mode is auto-detected from model capabilities, which makes the field
optional in the proxy OpenAPI spec. Regenerate the committed schema so the
dashboard types match: mode becomes optional and the description and default
JSDoc follow the spec, keeping the Check UI API Types Sync gate green

* refactor(bedrock): match all explicit route prefixes at path-segment boundary

Migrates the remaining substring route checks to the existing
_model_has_route_prefix helper so every explicit route token matches only as a
leading path segment, consistent with get_bedrock_route and the mantle route.
Covers _explicit_converse_route, _explicit_claude_platform_route,
_explicit_invoke_route, _explicit_agent_route, _explicit_agentcore_route,
_explicit_converse_like_route, _explicit_async_invoke_route and
_explicit_openai_route. This also stops invoke/ from substring-matching
async_invoke/. Route precedence and order are unchanged, and a note on the
segment invariant is added to the helper docstring

* test(bedrock): cover explicit route prefix segment matching

Exercises all eight migrated _explicit_*_route helpers (converse, converse_like,
invoke, async_invoke, agent, agentcore, claude_platform, openai) directly: each
matches its token as a leading path segment and rejects the token glued to a
preceding segment, so reverting any method to the old substring check fails the
suite. Also asserts invoke/ no longer matches async_invoke/ models, the concrete
improvement of the segment-boundary migration

* test(proxy): assert negative spend is allowed (one-time grant use-case)

Negative spend is intentionally permitted so admins can grant extra
allowance for the current budget period only, without raising the
recurring budget ceiling. Cover it explicitly in validate_finite_spend
and via the /user/update invalidation test.

* fix(google_genai): forward native generateContent top-level fields

Google's native generateContent REST body carries safetySettings, toolConfig,
cachedContent and labels at the top level as siblings of generationConfig. The
proxy's :generateContent endpoint spread them into agenerate_content as loose
kwargs and then dropped them, so callers had to wrap them in extra_body for them
to take effect; safetySettings, for instance, was silently ignored

The provider config now exposes the native top-level field names and
setup_generate_content_call collects whichever are present, merging them into the
outgoing request body through the existing extra_body merge so they reach Google
verbatim. An explicit extra_body still wins on conflict. The sync
generate_content_stream path now also forwards systemInstruction, matching the
other three entry points

Fixes #12671

Claude-Session: https://claude.ai/code/session_016MFtMXokCjT8u6mvyASudK

* fix(proxy): resolve env refs for DB-stored models

* fix(proxy): restrict DB env ref resolution

* fix(proxy): block team DB env ref resolution

* fix(lint): resolve ANN401/UP045/C901 strict-gate violations

- Replace Optional[X] with X | None (UP045) in 8 files
- Replace Any return/param types with concrete types or object (ANN401)
- Extract _make_api_key_auth_header helper to reduce get_anthropic_headers complexity below C901 threshold (17 → 14)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(anthropic): preserve x-api-key for custom endpoints; opt-in Bearer via prefix

Users who pass a key already prefixed with "Bearer " get Authorization: Bearer.
All other keys continue to use x-api-key, preserving backward compatibility with
custom api_base endpoints that expect x-api-key rather than Authorization.

Also consolidates get_auth_header to reuse _make_api_key_auth_header helper,
eliminating the duplicated custom-endpoint routing logic.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* revert(anthropic): restore Bearer routing for non-sk-ant- keys on custom api_base

The backwards-compat change broke existing tests that verify the intentional
Bearer-for-custom-base behavior (Fixes #30926). Restore original logic while
keeping the _make_api_key_auth_header helper for code deduplication.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(anthropic): gate Bearer-for-custom-base behind use_bearer_for_custom_base flag

Previously the auth-header switch from x-api-key to Authorization: Bearer
applied unconditionally for non-sk-ant- keys on a custom api_base, silently
breaking existing deployments that proxied to gateways expecting x-api-key.

Introduce use_bearer_for_custom_base: bool = False on _make_api_key_auth_header,
get_anthropic_headers, and get_auth_header. validate_environment reads it from
litellm_params so callers can opt in per-model without any API surface change.

Tests updated to pass use_bearer_for_custom_base=True where Bearer behavior is asserted.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(redis): apply namespace prefix in delete_cache and async_delete_cache (#29981)

DEL was the only Redis cache operation that skipped check_and_fix_namespace,
so it targeted the raw SHA256 hash (e.g. 3997c4...) rather than the
namespaced key (litellm:3997c4...). This caused two problems: a Redis NOPERM
error on deployments with an ACL restricting DEL to the litellm:* pattern,
and a silent no-op on all other deployments since the un-prefixed key was
never stored.

* style(anthropic): reformat common_utils.py with Black (--target-version py312)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix: preserve cache metadata and spend counters

* style: apply ruff format to streaming_iterator.py

* refactor: reduce complexity of usage/spend helpers to satisfy strict ruff gate

Extract Anthropic message_start cursor reset into
_reset_anthropic_cursor_completion_tokens and the cross-pod spend-counter
invalidation into _invalidate_user_spend_counter_if_changed, keeping both
_calculate_usage_per_chunk and _update_single_user_helper under the
max-complexity ceiling. Use builtin generics in the new signatures so no
new UP006 violations are introduced. Behavior unchanged.

---------

Co-authored-by: rupak-eng <rupakji99@gmail.com>
Co-authored-by: songkuan-zheng <252822057+songkuan-zheng@users.noreply.github.com>
Co-authored-by: songkuan-zheng <songkuan-zheng@users.noreply.github.com>
Co-authored-by: Kannan Priyadharshan <kpd2204@gmail.com>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Co-authored-by: Marco Georgaklis <mgeorgaklis@google.com>
Co-authored-by: Anjaiah Methuku <anjaiahspr@gmail.com>
Co-authored-by: Andrii Butko <booandrew23@gmail.com>
Co-authored-by: Kent <kingdooo@gmail.com>
Co-authored-by: kunal2002 <k.nayyar2002@gmail.com>
Co-authored-by: Ali Khan <alirazakhan.offi@gmail.com>
Co-authored-by: jesco-absolut <team@srswti.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Matt Hill <mhill@dataminr.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
2026-06-25 21:00:28 -07:00

3001 lines
108 KiB
Python

# Create server parameters for stdio connection
import os
import sys
import pytest
from unittest.mock import AsyncMock, MagicMock, patch
from contextlib import asynccontextmanager
sys.path.insert(
0, os.path.abspath("../../..")
) # Adds the parent directory to the system path
from litellm.proxy._experimental.mcp_server.mcp_server_manager import (
MCPServerManager,
MCPServer,
MCPTransport,
)
from litellm.proxy._types import LiteLLM_ObjectPermissionTable
from mcp.types import Tool as MCPTool, CallToolResult, ListToolsResult
from mcp.types import TextContent
mcp_server_manager = MCPServerManager()
@pytest.mark.asyncio
@pytest.mark.skip(reason="Local only test")
async def test_mcp_server_manager():
await mcp_server_manager.load_servers_from_config(
{
"zapier_mcp_server": {
"url": os.environ.get("ZAPIER_MCP_SERVER_URL"),
}
}
)
tools = await mcp_server_manager.list_tools()
print("TOOLS FROM MCP SERVER MANAGER== ", tools)
result = await mcp_server_manager.call_tool(
name="gmail_send_email", arguments={"body": "Test"}, proxy_logging_obj=None
)
print("RESULT FROM CALLING TOOL FROM MCP SERVER MANAGER== ", result)
@pytest.mark.asyncio
async def test_mcp_server_manager_https_server():
# Create mock tools and results
mock_tools = [
MCPTool(
name="gmail_send_email",
description="Send an email via Gmail",
inputSchema={
"type": "object",
"properties": {
"body": {"type": "string"},
"message": {"type": "string"},
"instructions": {"type": "string"},
},
"required": ["body"],
},
)
]
mock_result = CallToolResult(
content=[TextContent(type="text", text="Email sent successfully")],
isError=False,
)
# Create a mock MCPClient
mock_client = AsyncMock()
mock_client.list_tools = AsyncMock(return_value=mock_tools)
mock_client.call_tool = AsyncMock(return_value=mock_result)
# Mock the MCPClient constructor
def mock_client_constructor(*args, **kwargs):
return mock_client
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
await mcp_server_manager.load_servers_from_config(
{
"zapier_mcp_server": {
"url": "https://test-mcp-server.com/mcp",
"transport": MCPTransport.http,
}
}
)
allowed_server_ids = list(mcp_server_manager.get_registry().keys())
assert allowed_server_ids, "Expected registry to contain the configured server"
with patch.object(
mcp_server_manager,
"get_allowed_mcp_servers",
new=AsyncMock(return_value=allowed_server_ids),
):
tools = await mcp_server_manager.list_tools()
print("TOOLS FROM MCP SERVER MANAGER== ", tools)
# Verify tools were returned and properly prefixed
assert len(tools) == 1
# The server should use the server_name as prefix since no alias is provided
expected_prefix = "zapier_mcp_server"
assert tools[0].name == f"{expected_prefix}-gmail_send_email"
# Manually set up the tool mapping for the call_tool test
mcp_server_manager.tool_name_to_mcp_server_name_mapping["gmail_send_email"] = (
expected_prefix
)
mcp_server_manager.tool_name_to_mcp_server_name_mapping[
f"{expected_prefix}-gmail_send_email"
] = expected_prefix
result = await mcp_server_manager.call_tool(
server_name="zapier_mcp_server",
name=f"{expected_prefix}-gmail_send_email",
arguments={
"body": "Test",
"message": "Test",
"instructions": "Test",
},
proxy_logging_obj=None,
)
print("RESULT FROM CALLING TOOL FROM MCP SERVER MANAGER== ", result)
# Verify result
assert result.isError is False
assert len(result.content) == 1
assert isinstance(result.content[0], TextContent)
assert result.content[0].text == "Email sent successfully"
# Verify client methods were called
mock_client.list_tools.assert_called()
mock_client.call_tool.assert_called_once()
@pytest.mark.asyncio
async def test_mcp_http_transport_list_tools_mock():
"""Test HTTP transport list_tools functionality with mocked dependencies"""
# Create a fresh manager for testing
test_manager = MCPServerManager()
# Mock tools that should be returned
mock_tools = [
MCPTool(
name="gmail_send_email",
description="Send an email via Gmail",
inputSchema={
"type": "object",
"properties": {
"to": {"type": "string"},
"subject": {"type": "string"},
"body": {"type": "string"},
},
"required": ["to", "subject", "body"],
},
),
MCPTool(
name="calendar_create_event",
description="Create a calendar event",
inputSchema={
"type": "object",
"properties": {
"title": {"type": "string"},
"date": {"type": "string"},
"time": {"type": "string"},
},
"required": ["title", "date"],
},
),
]
# Create a mock MCPClient that returns our test tools
mock_client = AsyncMock()
mock_client.list_tools = AsyncMock(return_value=mock_tools)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
# Mock the MCPClient constructor to return our mock
def mock_client_constructor(*args, **kwargs):
return mock_client
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Load server config with HTTP transport
await test_manager.load_servers_from_config(
{
"test_http_server": {
"url": "https://test-mcp-server.com/mcp",
"transport": MCPTransport.http,
"description": "Test HTTP MCP Server",
}
}
)
# Call list_tools
allowed_server_ids = list(test_manager.get_registry().keys())
assert allowed_server_ids, "Expected registry to contain configured server"
with patch.object(
test_manager,
"get_allowed_mcp_servers",
new=AsyncMock(return_value=allowed_server_ids),
):
tools = await test_manager.list_tools()
# Assertions
assert len(tools) == 2
# The server should use the server_name as prefix since no alias is provided
expected_prefix = "test_http_server"
assert tools[0].name == f"{expected_prefix}-gmail_send_email"
assert tools[1].name == f"{expected_prefix}-calendar_create_event"
# Verify client methods were called
mock_client.list_tools.assert_called()
# Verify tool mapping was updated
expected_prefix = "test_http_server"
assert (
test_manager.tool_name_to_mcp_server_name_mapping[
f"{expected_prefix}-gmail_send_email"
]
== expected_prefix
)
assert (
test_manager.tool_name_to_mcp_server_name_mapping[
f"{expected_prefix}-calendar_create_event"
]
== expected_prefix
)
@pytest.mark.asyncio
async def test_mcp_http_transport_call_tool_mock():
"""Test HTTP transport call_tool functionality with mocked dependencies"""
# Create a fresh manager for testing
test_manager = MCPServerManager()
# Mock tool call result
mock_result = CallToolResult(
content=[
TextContent(type="text", text="Email sent successfully to test@example.com")
],
isError=False,
)
# Create a mock MCPClient that returns our test result
mock_client = AsyncMock()
mock_client.call_tool = AsyncMock(return_value=mock_result)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
# Mock the MCPClient constructor to return our mock
def mock_client_constructor(*args, **kwargs):
return mock_client
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Load server config with HTTP transport
await test_manager.load_servers_from_config(
{
"test_http_server": {
"url": "https://test-mcp-server.com/mcp",
"transport": MCPTransport.http,
"description": "Test HTTP MCP Server",
}
}
)
# Manually set up tool mapping (normally done by list_tools)
test_manager.tool_name_to_mcp_server_name_mapping["gmail_send_email"] = (
"test_http_server"
)
# Call the tool
result = await test_manager.call_tool(
server_name="test_http_server",
name="gmail_send_email",
arguments={
"to": "test@example.com",
"subject": "Test Subject",
"body": "Test email body",
},
proxy_logging_obj=None,
)
# Assertions
assert result.isError is False
assert len(result.content) == 1
# Type check before accessing text attribute
assert isinstance(result.content[0], TextContent)
assert result.content[0].text == "Email sent successfully to test@example.com"
# Verify client methods were called
mock_client.call_tool.assert_called_once()
@pytest.mark.asyncio
async def test_mcp_http_transport_call_tool_error_mock():
"""Test HTTP transport call_tool error handling with mocked dependencies"""
# Create a fresh manager for testing
test_manager = MCPServerManager()
# Mock tool call error result
mock_error_result = CallToolResult(
content=[TextContent(type="text", text="Error: Invalid email address")],
isError=True,
)
# Create a mock MCPClient that returns our test error result
mock_client = AsyncMock()
mock_client.call_tool = AsyncMock(return_value=mock_error_result)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
# Mock the MCPClient constructor to return our mock
def mock_client_constructor(*args, **kwargs):
return mock_client
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Load server config with HTTP transport
await test_manager.load_servers_from_config(
{
"test_http_server": {
"url": "https://test-mcp-server.com/mcp",
"transport": MCPTransport.http,
"description": "Test HTTP MCP Server",
}
}
)
# Manually set up tool mapping
test_manager.tool_name_to_mcp_server_name_mapping["gmail_send_email"] = (
"test_http_server"
)
# Call the tool with invalid data
result = await test_manager.call_tool(
server_name="test_http_server",
name="gmail_send_email",
arguments={"to": "invalid-email", "subject": "Test", "body": "Test"},
proxy_logging_obj=None,
)
# Assertions for error case
assert result.isError is True
assert len(result.content) == 1
# Type check before accessing text attribute
assert isinstance(result.content[0], TextContent)
assert "Error: Invalid email address" in result.content[0].text
# Verify client methods were called
mock_client.call_tool.assert_called_once()
@pytest.mark.asyncio
async def test_mcp_http_transport_tool_not_found():
"""Test calling a tool that doesn't exist"""
# Create a fresh manager for testing
test_manager = MCPServerManager()
# Load server config
await test_manager.load_servers_from_config(
{
"test_http_server": {
"url": "https://test-mcp-server.com/mcp",
"transport": MCPTransport.http,
"description": "Test HTTP MCP Server",
}
}
)
# Mapping populated for this server but not for the requested tool
test_manager.tool_name_to_mcp_server_name_mapping["gmail_send_email"] = (
"test_http_server"
)
# Try to call a tool that doesn't exist in mapping
with pytest.raises(ValueError, match="Tool nonexistent_tool not found"):
await test_manager.call_tool(
server_name="test_http_server",
name="nonexistent_tool",
arguments={"param": "value"},
proxy_logging_obj=None,
)
@pytest.mark.asyncio
async def test_streamable_http_mcp_handler_mock():
"""Test the streamable HTTP MCP handler functionality"""
# Mock streamable HTTP session managers and their methods
mock_session_manager_stateless = AsyncMock()
mock_session_manager_stateless.handle_request = AsyncMock()
mock_session_manager_stateful = AsyncMock()
mock_session_manager_stateful.handle_request = AsyncMock()
# Mock scope, receive, send with proper ASGI scope format
mock_scope = {
"type": "http",
"method": "POST",
"path": "/mcp",
"headers": [(b"content-type", b"application/json")],
"query_string": b"",
"server": ("localhost", 8000),
"scheme": "http",
}
mock_receive = AsyncMock(return_value={"body": b"{}", "more_body": False})
mock_send = AsyncMock()
# Mock extract_mcp_auth_context to bypass auth checks in the handler
mock_auth_context = (None, None, None, {}, {}, {})
with (
patch(
"litellm.proxy._experimental.mcp_server.server._SESSION_MANAGERS_INITIALIZED",
True,
),
patch(
"litellm.proxy._experimental.mcp_server.server.session_manager_stateless",
mock_session_manager_stateless,
),
patch(
"litellm.proxy._experimental.mcp_server.server.session_manager_stateful",
mock_session_manager_stateful,
),
patch(
"litellm.proxy._experimental.mcp_server.server.extract_mcp_auth_context",
AsyncMock(return_value=mock_auth_context),
),
patch(
"litellm.proxy._experimental.mcp_server.server.set_auth_context",
),
):
from litellm.proxy._experimental.mcp_server.server import (
handle_streamable_http_mcp,
)
# Call the handler
await handle_streamable_http_mcp(mock_scope, mock_receive, mock_send)
# Verify stateless session manager handle_request was called
mock_session_manager_stateless.handle_request.assert_called_once()
mock_session_manager_stateful.handle_request.assert_not_called()
@pytest.mark.asyncio
async def test_sse_mcp_handler_mock():
"""Test the SSE MCP handler functionality"""
from litellm.proxy._types import UserAPIKeyAuth
# Mock the SSE session manager and its methods
mock_sse_session_manager = AsyncMock()
mock_sse_session_manager.handle_request = AsyncMock()
# Mock scope, receive, send with proper ASGI scope format
mock_scope = {
"type": "http",
"method": "GET",
"path": "/mcp/sse",
"headers": [(b"accept", b"text/event-stream")],
"query_string": b"",
"server": ("localhost", 8000),
"scheme": "http",
}
mock_receive = AsyncMock()
mock_send = AsyncMock()
mock_auth_result = (
UserAPIKeyAuth(),
None,
None,
{},
{},
[],
)
with (
patch(
"litellm.proxy._experimental.mcp_server.server._SESSION_MANAGERS_INITIALIZED",
True,
),
patch(
"litellm.proxy._experimental.mcp_server.server.sse_session_manager",
mock_sse_session_manager,
),
patch(
"litellm.proxy._experimental.mcp_server.server.extract_mcp_auth_context",
new=AsyncMock(return_value=mock_auth_result),
),
patch(
"litellm.proxy._experimental.mcp_server.server.set_auth_context",
),
):
from litellm.proxy._experimental.mcp_server.server import handle_sse_mcp
# Call the handler
await handle_sse_mcp(mock_scope, mock_receive, mock_send)
# Verify SSE session manager handle_request was called
mock_sse_session_manager.handle_request.assert_called_once_with(
mock_scope, mock_receive, mock_send
)
@pytest.mark.asyncio
async def test_sse_mcp_handler_propagates_passthrough_401():
"""SSE handler must raise 401 + WWW-Authenticate when the upstream
pass-through probe rejects the client's bearer token, instead of letting
the SSE session start and silently return empty tool lists."""
from fastapi import HTTPException
from litellm.proxy._types import UserAPIKeyAuth
mock_scope = {
"type": "http",
"method": "GET",
"path": "/mcp/sse",
"headers": [(b"accept", b"text/event-stream")],
"query_string": b"",
"server": ("localhost", 8000),
"scheme": "http",
}
mock_receive = AsyncMock()
mock_send = AsyncMock()
mock_auth_result = (UserAPIKeyAuth(), None, None, {}, {}, [])
challenge = HTTPException(
status_code=401,
detail="Unauthorized",
headers={"WWW-Authenticate": "Bearer authorization_uri=https://example/"},
)
with (
patch(
"litellm.proxy._experimental.mcp_server.server._SESSION_MANAGERS_INITIALIZED",
True,
),
patch(
"litellm.proxy._experimental.mcp_server.server.sse_session_manager",
AsyncMock(),
),
patch(
"litellm.proxy._experimental.mcp_server.server.extract_mcp_auth_context",
new=AsyncMock(return_value=mock_auth_result),
),
patch(
"litellm.proxy._experimental.mcp_server.server.set_auth_context",
),
patch(
"litellm.proxy._experimental.mcp_server.server._raise_preemptive_401_for_unauthenticated_servers",
new=AsyncMock(),
),
patch(
"litellm.proxy._experimental.mcp_server.server._check_passthrough_upstream_auth",
new=AsyncMock(side_effect=challenge),
),
):
from litellm.proxy._experimental.mcp_server.server import handle_sse_mcp
with pytest.raises(HTTPException) as excinfo:
await handle_sse_mcp(mock_scope, mock_receive, mock_send)
assert excinfo.value.status_code == 401
assert excinfo.value.headers and "WWW-Authenticate" in excinfo.value.headers
def test_generate_stable_server_id():
"""
Test the _generate_stable_server_id method to ensure hash stability across releases.
This test verifies that:
1. The same inputs always produce the same hash output
2. Different inputs produce different hash outputs
3. The hash format is consistent (32 character hex string)
4. Edge cases work correctly (None auth_type)
IMPORTANT: If this test fails, it means the hashing algorithm has changed
and will break backwards compatibility with existing server IDs!
"""
manager = MCPServerManager()
# Test Case 1: Basic functionality with known inputs
# These expected values MUST remain stable across releases
test_cases = [
{
"params": {
"server_name": "zapier_mcp_server",
"url": "https://actions.zapier.com/mcp/sse",
"transport": "sse",
"auth_type": "api_key",
},
"expected_hash": "8d5c9f8a12e3b7c4f6a2d8e1b5c9f2a4",
},
{
"params": {
"server_name": "google_drive_mcp_server",
"url": "https://drive.google.com/mcp/http",
"transport": "http",
"auth_type": None,
},
"expected_hash": "7a4b2e8f3c1d9e6b5a7c8f2d4e1b9c6a",
},
{
"params": {
"server_name": "local_test_server",
"url": "http://localhost:8080/mcp",
"transport": "http",
"auth_type": "basic",
},
"expected_hash": "2f1e8d7c6b5a4e3f2d1c9b8a7e6f5d4c",
},
]
# Test that our known inputs produce expected hash values
for test_case in test_cases:
result = manager._generate_stable_server_id(**test_case["params"])
# For now, just verify the format and stability, not exact hash
# (since we need to first run to see what the actual hashes are)
assert len(result) == 32, f"Hash should be 32 characters, got {len(result)}"
assert result.isalnum(), f"Hash should be alphanumeric, got: {result}"
assert result.islower(), f"Hash should be lowercase, got: {result}"
# Test stability - same inputs should always produce same output
result2 = manager._generate_stable_server_id(**test_case["params"])
assert (
result == result2
), f"Hash should be stable for same inputs: {result} != {result2}"
# Test Case 2: Different inputs produce different outputs
base_params = {
"server_name": "test_server",
"url": "https://test.com/mcp",
"transport": "sse",
"auth_type": "api_key",
}
base_hash = manager._generate_stable_server_id(**base_params)
# Change each parameter and verify hash changes
variations = [
{"server_name": "different_server"},
{"url": "https://different.com/mcp"},
{"transport": "http"},
{"auth_type": "basic"},
{"auth_type": None},
]
for variation in variations:
modified_params = {**base_params, **variation}
modified_hash = manager._generate_stable_server_id(**modified_params)
assert (
modified_hash != base_hash
), f"Different params should produce different hash: {variation}"
assert (
len(modified_hash) == 32
), f"Modified hash should be 32 characters: {variation}"
# Test Case 3: Edge case with None auth_type
params_with_none = {
"server_name": "test_server",
"url": "https://test.com/mcp",
"transport": "sse",
"auth_type": None,
}
params_with_empty = {
"server_name": "test_server",
"url": "https://test.com/mcp",
"transport": "sse",
"auth_type": "",
}
hash_none = manager._generate_stable_server_id(**params_with_none)
hash_empty = manager._generate_stable_server_id(**params_with_empty)
# None and empty string should produce the same hash (both become empty string)
assert (
hash_none == hash_empty
), "None auth_type should be equivalent to empty string"
# Test Case 4: Real-world example hashes that must remain stable
# These are based on common configurations and MUST NOT CHANGE
zapier_sse_hash = manager._generate_stable_server_id(
server_name="zapier_mcp_server",
url="https://actions.zapier.com/mcp/sk-ak-example/sse",
transport="sse",
auth_type="api_key",
)
github_http_hash = manager._generate_stable_server_id(
server_name="github_mcp_server",
url="https://api.github.com/mcp/http",
transport="http",
auth_type=None,
)
# These should be deterministic - same call should produce same result
assert zapier_sse_hash == manager._generate_stable_server_id(
server_name="zapier_mcp_server",
url="https://actions.zapier.com/mcp/sk-ak-example/sse",
transport="sse",
auth_type="api_key",
)
assert github_http_hash == manager._generate_stable_server_id(
server_name="github_mcp_server",
url="https://api.github.com/mcp/http",
transport="http",
auth_type=None,
)
# Verify format
assert len(zapier_sse_hash) == 32
assert len(github_http_hash) == 32
assert zapier_sse_hash != github_http_hash
@pytest.mark.asyncio
async def test_list_tools_rest_api_server_not_found():
"""Test the list_tools REST API when server is not found"""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
list_tool_rest_api,
global_mcp_server_manager,
)
from fastapi import Query
from litellm.proxy._types import UserAPIKeyAuth
# Mock UserAPIKeyAuth with explicit permission to access the requested server id
mock_user_auth = UserAPIKeyAuth(
api_key="test",
user_id="test",
object_permission=LiteLLM_ObjectPermissionTable(
object_permission_id="dummy",
mcp_servers=["non_existent_server_id"],
),
)
# Mock request with proper client attribute (internal IP for no filtering)
mock_request = MagicMock()
mock_request.headers = {}
mock_request.client = MagicMock()
mock_request.client.host = "127.0.0.1" # Internal IP to bypass IP filtering
# Mock the global_mcp_server_manager to allow the server ID but return None for the server
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints.global_mcp_server_manager"
) as mock_manager:
# Allow the server ID in permissions
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["non_existent_server_id"]
)
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
# Return None when trying to get the server (server doesn't exist)
mock_manager.get_mcp_server_by_id = MagicMock(return_value=None)
# Test with non-existent server ID
response = await list_tool_rest_api(
request=mock_request,
server_id="non_existent_server_id",
user_api_key_dict=mock_user_auth,
)
assert isinstance(response, dict)
assert response["tools"] == []
assert response["error"] == "server_not_found"
assert "Server with id non_existent_server_id not found" in response["message"]
@pytest.mark.asyncio
async def test_list_tools_rest_api_success():
"""Test the list_tools REST API successful case"""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
list_tool_rest_api,
)
from litellm.proxy._experimental.mcp_server.server import (
ListMCPToolsRestAPIResponseObject,
)
from fastapi import Query
from litellm.proxy._types import UserAPIKeyAuth
# Mock successful tools
mock_tools = [
ListMCPToolsRestAPIResponseObject(
name="test_tool",
description="A test tool",
inputSchema={"type": "object"},
mcp_info={"server_name": "test_server"},
)
]
# Create a mock server
mock_server = MagicMock()
mock_server.server_id = "test-server-123"
mock_server.alias = "test_server"
mock_server.name = "test_server"
mock_server.mcp_info = {"server_name": "test_server"}
# Mock UserAPIKeyAuth
mock_user_auth = UserAPIKeyAuth(
api_key="test",
user_id="test",
object_permission=LiteLLM_ObjectPermissionTable(
object_permission_id="dummy",
mcp_servers=["test-server-123"],
),
)
# Mock request with proper client attribute (internal IP for no filtering)
mock_request = MagicMock()
mock_request.headers = {}
mock_request.client = MagicMock()
mock_request.client.host = "127.0.0.1" # Internal IP to bypass IP filtering
# Mock the global_mcp_server_manager
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints.global_mcp_server_manager"
) as mock_manager:
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["test-server-123"]
)
mock_manager.get_mcp_server_by_id = MagicMock(return_value=mock_server)
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
# Mock the _get_tools_for_single_server function
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints._get_tools_for_single_server"
) as mock_get_tools:
mock_get_tools.return_value = mock_tools
# Test successful case
response = await list_tool_rest_api(
request=mock_request,
server_id="test-server-123",
user_api_key_dict=mock_user_auth,
)
assert isinstance(response, dict)
assert len(response["tools"]) == 1
assert response["tools"][0].name == "test_tool"
@pytest.mark.asyncio
async def test_get_tools_from_mcp_servers():
"""Test _get_tools_from_mcp_servers function with both specific and no server filters"""
from litellm.proxy._experimental.mcp_server.server import (
_get_tools_from_mcp_servers,
)
from litellm.proxy._types import UserAPIKeyAuth
from litellm.proxy._experimental.mcp_server.mcp_server_manager import (
MCPServer,
MCPTransport,
)
# Mock data
mock_user_auth = UserAPIKeyAuth(api_key="test_key", user_id="test_user")
mock_auth_header = "Bearer test_token"
mock_server_1 = MCPServer(
server_id="server1_id",
name="server1",
server_name="server1",
url="http://test1.com",
transport=MCPTransport.http,
)
mock_server_2 = MCPServer(
server_id="server2_id",
name="server2",
server_name="server2",
url="http://test2.com",
transport=MCPTransport.http,
)
mock_server_3 = MCPServer(
server_id="server3_id",
name="server3",
server_name="server3",
url="http://test3.com",
transport=MCPTransport.http,
access_groups=["group-a"],
)
mock_tool_1 = MCPTool(name="tool1", description="test tool 1", inputSchema={})
mock_tool_2 = MCPTool(name="tool2", description="test tool 2", inputSchema={})
# Test Case 1: With specific MCP servers
try:
# Mock the necessary methods
def mock_get_server_by_id(server_id):
if server_id == "server1_id":
return mock_server_1
elif server_id == "server2_id":
return mock_server_2
elif server_id == "server3_id":
return mock_server_3
return None
# Create a mock manager
mock_manager = AsyncMock()
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["server1_id", "server2_id"]
)
mock_manager.get_mcp_server_by_id = lambda server_id: (
mock_server_1 if server_id == "server1_id" else mock_server_2
)
mock_manager._get_tools_from_server = AsyncMock(return_value=[mock_tool_1])
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
with patch(
"litellm.proxy._experimental.mcp_server.server.global_mcp_server_manager",
mock_manager,
):
# Test with specific servers
result = await _get_tools_from_mcp_servers(
user_api_key_auth=mock_user_auth,
mcp_auth_header=mock_auth_header,
mcp_servers=["server1"],
)
assert len(result) == 1, "Should only return tools from server1"
assert result[0].name == "tool1", "Should return tool from server1"
# Test Case 2: Without specific MCP servers
# Create a different mock manager for the second test case
mock_manager_2 = AsyncMock()
mock_manager_2.get_allowed_mcp_servers = AsyncMock(
return_value=["server1_id", "server2_id"]
)
mock_manager_2.get_mcp_server_by_id = lambda server_id: (
mock_server_1 if server_id == "server1_id" else mock_server_2
)
async def mock_get_tools_side_effect(
server,
mcp_auth_header=None,
extra_headers=None,
add_prefix=False,
raw_headers=None,
user_api_key_auth=None,
):
if server.server_id == "server1_id":
return [mock_tool_1]
return [mock_tool_2]
mock_manager_2._get_tools_from_server = AsyncMock(
side_effect=mock_get_tools_side_effect
)
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager_2.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
with patch(
"litellm.proxy._experimental.mcp_server.server.global_mcp_server_manager",
mock_manager_2,
):
result = await _get_tools_from_mcp_servers(
user_api_key_auth=mock_user_auth,
mcp_auth_header=mock_auth_header,
mcp_servers=None,
)
assert len(result) == 2, "Should return tools from all servers"
assert (
result[0].name == "tool1" and result[1].name == "tool2"
), "Should return tools from all servers"
#
# Test Case 3: With specific MCP servers and access groups
# Create a mock manager
mock_manager = AsyncMock()
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["server1_id", "server2_id", "server3_id"]
)
mock_manager.get_mcp_server_by_id = lambda server_id: (
mock_server_1
if server_id == "server1_id"
else (mock_server_2 if server_id == "server2_id" else mock_server_3)
)
mock_manager._get_tools_from_server = AsyncMock(return_value=[mock_tool_1])
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
with patch(
"litellm.proxy._experimental.mcp_server.server.global_mcp_server_manager",
mock_manager,
):
with patch(
"litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp.MCPRequestHandler._get_mcp_servers_from_access_groups",
AsyncMock(return_value=["server3_id"]),
):
# Test with specific servers
result = await _get_tools_from_mcp_servers(
user_api_key_auth=mock_user_auth,
mcp_auth_header=mock_auth_header,
mcp_servers=["group-a"],
)
assert len(result) == 1, "Should only return tools from server3"
assert result[0].name == "tool1", "Should return tool from server1"
except AssertionError as e:
pytest.fail(f"Test failed: {str(e)}")
except Exception as e:
pytest.fail(f"Unexpected error in tests: {str(e)}")
@pytest.mark.asyncio
async def test_list_tools_only_returns_allowed_servers(monkeypatch):
"""
Test that list_tools only returns tools from servers allowed for the user.
"""
test_manager = MCPServerManager()
# Setup two servers in the config
await test_manager.load_servers_from_config(
{
"server_a": {
"url": "https://server-a.com/mcp",
"transport": MCPTransport.http,
"description": "Server A",
},
"server_b": {
"url": "https://server-b.com/mcp",
"transport": MCPTransport.http,
"description": "Server B",
},
}
)
# Patch get_allowed_mcp_servers to only allow server_a
async def mock_get_allowed_mcp_servers(self, user_api_key_auth=None):
return [
list(test_manager.get_registry().keys())[0]
] # Only first server (server_a)
monkeypatch.setattr(
MCPServerManager, "get_allowed_mcp_servers", mock_get_allowed_mcp_servers
)
# Mock tools for each server
mock_tools_a = [
MCPTool(
name="send_email",
description="Send an email via Server A",
inputSchema={"type": "object"},
)
]
mock_tools_b = [
MCPTool(
name="create_event",
description="Create an event via Server B",
inputSchema={"type": "object"},
)
]
# Patch MCPClient to return different tools for each server
def mock_client_constructor(*args, **kwargs):
mock_client = AsyncMock()
# Return tools based on server URL
if kwargs.get("server_url") == "https://server-a.com/mcp":
mock_client.list_tools = AsyncMock(return_value=mock_tools_a)
else:
mock_client.list_tools = AsyncMock(return_value=mock_tools_b)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
return mock_client
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Call list_tools
from litellm.proxy._types import UserAPIKeyAuth
tools = await test_manager.list_tools(user_api_key_auth=UserAPIKeyAuth())
# Should only return tools from server_a
assert len(tools) == 1
# The server should use the server_name as prefix since no alias is provided
expected_prefix = "server_a"
assert tools[0].name.startswith(f"{expected_prefix}-")
@pytest.mark.asyncio
async def test_mcp_server_manager_access_groups_from_config():
"""
Test that access_groups are loaded from config and can be resolved.
"""
test_manager = MCPServerManager()
await test_manager.load_servers_from_config(
{
"config_server": {
"url": "https://config-mcp-server.com/mcp",
"transport": MCPTransport.http,
"access_groups": ["group-a", "group-b"],
},
"other_server": {
"url": "https://other-mcp-server.com/mcp",
"transport": MCPTransport.http,
"access_groups": ["group-b", "group-c"],
},
}
)
# Check that access_groups are loaded
config_server = next(
(
s
for s in test_manager.config_mcp_servers.values()
if s.name == "config_server"
),
None,
)
assert config_server is not None
assert set(config_server.access_groups) == {"group-a", "group-b"}
# Check that the lookup logic finds the correct server ids
from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import (
MCPRequestHandler,
)
# Patch global_mcp_server_manager for this test and restore afterwards to
# avoid leaking state into other tests (e.g. the proxy MCP e2e suite).
import litellm.proxy._experimental.mcp_server.mcp_server_manager as mcp_server_manager_mod
original_manager = mcp_server_manager_mod.global_mcp_server_manager
mcp_server_manager_mod.global_mcp_server_manager = test_manager
try:
# Should find config_server for group-a, both for group-b, other_server for group-c
import asyncio
server_ids_a = await MCPRequestHandler._get_mcp_servers_from_access_groups(
["group-a"]
)
server_ids_b = await MCPRequestHandler._get_mcp_servers_from_access_groups(
["group-b"]
)
server_ids_c = await MCPRequestHandler._get_mcp_servers_from_access_groups(
["group-c"]
)
assert any(config_server.server_id == sid for sid in server_ids_a)
assert set(server_ids_b) == set(
[
s.server_id
for s in test_manager.config_mcp_servers.values()
if "group-b" in s.access_groups
]
)
assert any(
s.name == "other_server" and s.server_id in server_ids_c
for s in test_manager.config_mcp_servers.values()
)
finally:
mcp_server_manager_mod.global_mcp_server_manager = original_manager
@pytest.mark.asyncio
async def test_mcp_server_manager_config_integration_with_database():
"""
Test that config-based servers properly integrate with database servers,
specifically testing access_groups and description fields.
"""
import datetime
from litellm.proxy._types import LiteLLM_MCPServerTable
test_manager = MCPServerManager()
# Test 1: Load config with access_groups and description
await test_manager.load_servers_from_config(
{
"config_server_with_groups": {
"url": "https://config-server.com/mcp",
"transport": MCPTransport.http,
"description": "Test config server",
"access_groups": ["fr_staff", "admin"],
}
}
)
# Verify config server has correct access_groups
config_servers = test_manager.config_mcp_servers
assert len(config_servers) == 1
config_server = next(iter(config_servers.values()))
assert config_server.access_groups == ["fr_staff", "admin"]
assert config_server.mcp_info["description"] == "Test config server"
# Test 2: Create a database server record and test add_update_server method
db_server = LiteLLM_MCPServerTable(
server_id="db-server-123",
server_name="database-server",
url="https://db-server.com/mcp",
transport="http",
auth_type="none",
description="Database server description",
created_at=datetime.datetime.now(),
updated_at=datetime.datetime.now(),
mcp_access_groups=["db_group", "test_group"],
)
# Test the add_update_server method (this tests our fix)
await test_manager.add_server(db_server)
# Verify the server was added with correct access_groups
registry = test_manager.get_registry()
assert "db-server-123" in registry
db_server_in_registry = registry["db-server-123"]
assert db_server_in_registry.access_groups == ["db_group", "test_group"]
assert db_server_in_registry.server_name == "database-server"
# Test 3: Test config server conversion to LiteLLM_MCPServerTable format
# This tests that config servers are properly converted with access_groups and description fields
# Mock user auth to get all servers
from litellm.proxy._types import UserAPIKeyAuth
mock_user_auth = UserAPIKeyAuth(user_role="proxy_admin")
# Mock the get_allowed_mcp_servers to return only config server IDs
# (to avoid database dependency in this test)
async def mock_get_allowed_servers(user_auth=None):
config_server_ids = list(test_manager.config_mcp_servers.keys())
return config_server_ids
test_manager.get_allowed_mcp_servers = mock_get_allowed_servers
# Mock health_check_server to avoid real network calls that timeout
async def mock_health_check(server_id: str, mcp_auth_header=None):
server = test_manager.get_mcp_server_by_id(server_id)
if not server:
return None
return LiteLLM_MCPServerTable(
server_id=server_id,
server_name=server.name,
url=server.url,
transport=server.transport,
description=server.mcp_info.get("description") if server.mcp_info else None,
mcp_access_groups=server.access_groups,
status="healthy",
last_health_check=datetime.datetime.now(),
mcp_info=server.mcp_info,
)
test_manager.health_check_server = mock_health_check
# Test the method (this tests our second fix)
servers_list = await test_manager.get_all_mcp_servers_with_health_and_teams(
user_api_key_auth=mock_user_auth
)
# Verify we have the config server properly converted
assert len(servers_list) == 1
# Find the config server in the list
config_server_in_list = servers_list[0]
assert config_server_in_list.server_name == "config_server_with_groups"
assert config_server_in_list.mcp_access_groups == ["fr_staff", "admin"]
assert config_server_in_list.description == "Test config server"
# Verify the mcp_info is also correct
assert config_server_in_list.mcp_info["description"] == "Test config server"
assert config_server_in_list.mcp_info["server_name"] == "config_server_with_groups"
# Tests for Server Alias Functionality
def test_get_server_prefix_with_alias():
"""
Test that get_server_prefix returns alias when present.
"""
from litellm.proxy._experimental.mcp_server.utils import get_server_prefix
# Create a mock server with alias
mock_server = MagicMock()
mock_server.alias = "my_alias"
mock_server.server_name = "My Server Name"
mock_server.server_id = "server-123"
prefix = get_server_prefix(mock_server)
assert prefix == "my_alias"
def test_get_server_prefix_without_alias():
"""
Test that get_server_prefix falls back to server_name when alias is not present.
"""
from litellm.proxy._experimental.mcp_server.utils import get_server_prefix
# Create a mock server without alias
mock_server = MagicMock()
mock_server.alias = None
mock_server.server_name = "My Server Name"
mock_server.server_id = "server-123"
prefix = get_server_prefix(mock_server)
assert prefix == "My Server Name"
def test_get_server_prefix_fallback_to_server_id():
"""
Test that get_server_prefix falls back to server_id when neither alias nor server_name are present.
"""
from litellm.proxy._experimental.mcp_server.utils import get_server_prefix
# Create a mock server without alias or server_name
mock_server = MagicMock()
mock_server.alias = None
mock_server.server_name = None
mock_server.server_id = "server-123"
prefix = get_server_prefix(mock_server)
assert prefix == "server-123"
def test_get_server_prefix_empty_strings():
"""
Test that get_server_prefix handles empty strings correctly.
"""
from litellm.proxy._experimental.mcp_server.utils import get_server_prefix
# Create a mock server with empty strings
mock_server = MagicMock()
mock_server.alias = ""
mock_server.server_name = ""
mock_server.server_id = "server-123"
prefix = get_server_prefix(mock_server)
assert prefix == "server-123"
@pytest.mark.asyncio
async def test_mcp_server_manager_alias_tool_prefixing():
"""
Test that MCP server manager uses alias for tool prefixing when available.
"""
test_manager = MCPServerManager()
# Create a mock server with alias
mock_server = MCPServer(
server_id="test-server-123",
name="test_server",
alias="my_alias",
server_name="Test Server",
url="https://test-server.com/mcp",
transport=MCPTransport.http,
)
# Add server to registry
test_manager.registry["test-server-123"] = mock_server
# Mock tools
mock_tools = [
MCPTool(
name="send_email",
description="Send an email",
inputSchema={"type": "object"},
)
]
# Mock MCPClient
mock_client = AsyncMock()
mock_client.list_tools = AsyncMock(return_value=mock_tools)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
def mock_client_constructor(*args, **kwargs):
return mock_client
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Get tools from server
tools = await test_manager._get_tools_from_server(mock_server)
# Verify tool is prefixed with alias
assert len(tools) == 1
assert tools[0].name == "my_alias-send_email"
# Verify mapping is updated correctly
assert (
test_manager.tool_name_to_mcp_server_name_mapping["send_email"]
== "my_alias"
)
assert (
test_manager.tool_name_to_mcp_server_name_mapping["my_alias-send_email"]
== "my_alias"
)
@pytest.mark.asyncio
async def test_mcp_server_manager_server_name_tool_prefixing():
"""
Test that MCP server manager falls back to server_name for tool prefixing when alias is not available.
"""
test_manager = MCPServerManager()
# Create a mock server without alias
mock_server = MCPServer(
server_id="test-server-123",
name="test_server",
alias=None,
server_name="Test Server",
url="https://test-server.com/mcp",
transport=MCPTransport.http,
)
# Add server to registry
test_manager.registry["test-server-123"] = mock_server
# Mock tools
mock_tools = [
MCPTool(
name="send_email",
description="Send an email",
inputSchema={"type": "object"},
)
]
# Mock MCPClient
mock_client = AsyncMock()
mock_client.list_tools = AsyncMock(return_value=mock_tools)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
def mock_client_constructor(*args, **kwargs):
return mock_client
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Get tools from server
tools = await test_manager._get_tools_from_server(mock_server)
# Verify tool is prefixed with server_name (normalized)
assert len(tools) == 1
assert tools[0].name == "Test_Server-send_email"
# Verify mapping is updated correctly
assert (
test_manager.tool_name_to_mcp_server_name_mapping["send_email"]
== "Test Server"
)
assert (
test_manager.tool_name_to_mcp_server_name_mapping["Test_Server-send_email"]
== "Test Server"
)
@pytest.mark.asyncio
async def test_mcp_server_manager_server_id_tool_prefixing():
"""
Test that MCP server manager falls back to server_id for tool prefixing when neither alias nor server_name are available.
"""
test_manager = MCPServerManager()
# Create a mock server without alias or server_name
mock_server = MCPServer(
server_id="test-server-123",
name="test_server",
alias=None,
server_name=None,
url="https://test-server.com/mcp",
transport=MCPTransport.http,
)
# Add server to registry
test_manager.registry["test-server-123"] = mock_server
# Mock tools
mock_tools = [
MCPTool(
name="send_email",
description="Send an email",
inputSchema={"type": "object"},
)
]
# Mock MCPClient
mock_client = AsyncMock()
mock_client.list_tools = AsyncMock(return_value=mock_tools)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
def mock_client_constructor(*args, **kwargs):
return mock_client
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Get tools from server
tools = await test_manager._get_tools_from_server(mock_server)
# Verify tool is prefixed with server_id
assert len(tools) == 1
assert tools[0].name == "test-server-123-send_email"
# Verify mapping is updated correctly
assert (
test_manager.tool_name_to_mcp_server_name_mapping["send_email"]
== "test-server-123"
)
assert (
test_manager.tool_name_to_mcp_server_name_mapping[
"test-server-123-send_email"
]
== "test-server-123"
)
@pytest.mark.asyncio
async def test_add_update_server_with_alias():
"""
Test that add_update_server correctly handles servers with alias.
"""
test_manager = MCPServerManager()
# Create a mock LiteLLM_MCPServerTable with alias
mock_mcp_server = MagicMock()
mock_mcp_server.server_id = "test-server-123"
mock_mcp_server.alias = "my_alias"
mock_mcp_server.server_name = "Test Server"
mock_mcp_server.url = "https://test-server.com/mcp"
mock_mcp_server.transport = MCPTransport.http
mock_mcp_server.auth_type = None
mock_mcp_server.credentials = {}
mock_mcp_server.description = "Test server description"
mock_mcp_server.mcp_info = {}
mock_mcp_server.static_headers = {}
mock_mcp_server.command = None
mock_mcp_server.args = []
mock_mcp_server.env = None
mock_mcp_server.spec_path = None
# OAuth fields - set explicitly to None to avoid MagicMock objects
mock_mcp_server.client_id = None
mock_mcp_server.client_secret = None
mock_mcp_server.authorization_url = None
mock_mcp_server.registration_url = None
mock_mcp_server.token_url = None
mock_mcp_server.oauth2_flow = None
# Additional fields used by build_mcp_server_from_table
mock_mcp_server.extra_headers = None
mock_mcp_server.allow_all_keys = False
mock_mcp_server.available_on_public_internet = True
mock_mcp_server.mcp_access_groups = None
mock_mcp_server.allowed_tools = None
mock_mcp_server.disallowed_tools = None
mock_mcp_server.tool_name_to_display_name = None
mock_mcp_server.tool_name_to_description = None
mock_mcp_server.is_byok = False
mock_mcp_server.byok_description = None
mock_mcp_server.byok_api_key_help_url = None
mock_mcp_server.created_at = None
mock_mcp_server.updated_at = None
mock_mcp_server.instructions = None
mock_mcp_server.source_url = None
mock_mcp_server.approval_status = "active"
# Add server to manager
await test_manager.add_server(mock_mcp_server)
# Verify server was added with correct name (should use alias)
assert "test-server-123" in test_manager.registry
added_server = test_manager.registry["test-server-123"]
assert added_server.name == "my_alias"
assert added_server.alias == "my_alias"
assert added_server.server_name == "Test Server"
@pytest.mark.asyncio
async def test_add_update_server_without_alias():
"""
Test that add_update_server correctly handles servers without alias.
"""
test_manager = MCPServerManager()
# Create a mock LiteLLM_MCPServerTable without alias
mock_mcp_server = MagicMock()
mock_mcp_server.server_id = "test-server-123"
mock_mcp_server.alias = None
mock_mcp_server.server_name = "Test Server"
mock_mcp_server.url = "https://test-server.com/mcp"
mock_mcp_server.transport = MCPTransport.http
mock_mcp_server.auth_type = None
mock_mcp_server.credentials = {}
mock_mcp_server.description = "Test server description"
mock_mcp_server.mcp_info = {}
mock_mcp_server.static_headers = {}
mock_mcp_server.command = None
mock_mcp_server.args = []
mock_mcp_server.env = None
mock_mcp_server.spec_path = None
# OAuth fields - set explicitly to None to avoid MagicMock objects
mock_mcp_server.client_id = None
mock_mcp_server.client_secret = None
mock_mcp_server.authorization_url = None
mock_mcp_server.registration_url = None
mock_mcp_server.token_url = None
mock_mcp_server.oauth2_flow = None
# Additional fields used by build_mcp_server_from_table
mock_mcp_server.extra_headers = None
mock_mcp_server.allow_all_keys = False
mock_mcp_server.available_on_public_internet = True
mock_mcp_server.mcp_access_groups = None
mock_mcp_server.allowed_tools = None
mock_mcp_server.disallowed_tools = None
mock_mcp_server.tool_name_to_display_name = None
mock_mcp_server.tool_name_to_description = None
mock_mcp_server.is_byok = False
mock_mcp_server.byok_description = None
mock_mcp_server.byok_api_key_help_url = None
mock_mcp_server.created_at = None
mock_mcp_server.updated_at = None
mock_mcp_server.instructions = None
mock_mcp_server.source_url = None
mock_mcp_server.approval_status = "active"
# Add server to manager
await test_manager.add_server(mock_mcp_server)
# Verify server was added with correct name (should use server_name)
assert "test-server-123" in test_manager.registry
added_server = test_manager.registry["test-server-123"]
assert added_server.name == "Test Server"
assert added_server.alias is None
assert added_server.server_name == "Test Server"
@pytest.mark.asyncio
async def test_add_update_server_fallback_to_server_id():
"""
Test that add_update_server falls back to server_id when neither alias nor server_name are available.
"""
test_manager = MCPServerManager()
# Create a mock LiteLLM_MCPServerTable without alias or server_name
mock_mcp_server = MagicMock()
mock_mcp_server.server_id = "test-server-123"
mock_mcp_server.alias = None
mock_mcp_server.server_name = None
mock_mcp_server.url = "https://test-server.com/mcp"
mock_mcp_server.transport = MCPTransport.http
mock_mcp_server.auth_type = None
mock_mcp_server.credentials = {}
mock_mcp_server.description = "Test server description"
mock_mcp_server.mcp_info = {}
mock_mcp_server.static_headers = {}
mock_mcp_server.command = None
mock_mcp_server.args = []
mock_mcp_server.env = None
mock_mcp_server.spec_path = None
# OAuth fields - set explicitly to None to avoid MagicMock objects
mock_mcp_server.client_id = None
mock_mcp_server.client_secret = None
mock_mcp_server.authorization_url = None
mock_mcp_server.registration_url = None
mock_mcp_server.token_url = None
mock_mcp_server.oauth2_flow = None
# Additional fields used by build_mcp_server_from_table - set explicitly
# to avoid MagicMock objects being passed to Pydantic MCPServer constructor
mock_mcp_server.extra_headers = None
mock_mcp_server.allow_all_keys = False
mock_mcp_server.available_on_public_internet = True
mock_mcp_server.mcp_access_groups = None
mock_mcp_server.allowed_tools = None
mock_mcp_server.disallowed_tools = None
mock_mcp_server.tool_name_to_display_name = None
mock_mcp_server.tool_name_to_description = None
mock_mcp_server.is_byok = False
mock_mcp_server.byok_description = None
mock_mcp_server.byok_api_key_help_url = None
mock_mcp_server.created_at = None
mock_mcp_server.updated_at = None
mock_mcp_server.instructions = None
mock_mcp_server.source_url = None
mock_mcp_server.approval_status = "active"
# Add server to manager
await test_manager.add_server(mock_mcp_server)
# Verify server was added with correct name (should use server_id)
assert "test-server-123" in test_manager.registry
added_server = test_manager.registry["test-server-123"]
assert added_server.name == "test-server-123"
assert added_server.alias is None
assert added_server.server_name is None
def test_normalize_server_name():
"""
Test that normalize_server_name correctly replaces spaces with underscores.
"""
from litellm.proxy._experimental.mcp_server.utils import normalize_server_name
# Test basic space replacement
assert normalize_server_name("My Server Name") == "My_Server_Name"
# Test multiple consecutive spaces
assert normalize_server_name("My Server Name") == "My__Server___Name"
# Test no spaces
assert normalize_server_name("MyServerName") == "MyServerName"
# Test empty string
assert normalize_server_name("") == ""
# Test string with only spaces
assert normalize_server_name(" ") == "___"
def test_add_server_prefix_to_name():
"""Ensure add_server_prefix_to_name correctly formats resource names."""
from litellm.proxy._experimental.mcp_server.utils import add_server_prefix_to_name
# Test basic prefixing
result = add_server_prefix_to_name("send_email", "My Server")
assert result == "My_Server-send_email"
# Test with server name that already has underscores
result = add_server_prefix_to_name("create_event", "my_server")
assert result == "my_server-create_event"
# Test with empty name
result = add_server_prefix_to_name("", "My Server")
assert result == "My_Server-"
# Test with empty server name
result = add_server_prefix_to_name("send_email", "")
assert result == "-send_email"
def test_get_server_auth_header_with_alias():
"""Test _get_server_auth_header function with server alias."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
_get_server_auth_header,
)
# Create a mock server with alias
mock_server = MagicMock()
mock_server.alias = "zapier"
mock_server.server_name = "zapier_server"
# Test with server-specific auth headers
mcp_server_auth_headers = {
"zapier": "Bearer zapier_token",
"slack": "Bearer slack_token",
}
mcp_auth_header = "Bearer default_token"
result = _get_server_auth_header(
mock_server, mcp_server_auth_headers, mcp_auth_header
)
assert result == "Bearer zapier_token"
# Test case-insensitive matching
mcp_server_auth_headers = {
"ZAPIER": "Bearer zapier_token_upper",
"slack": "Bearer slack_token",
}
result = _get_server_auth_header(
mock_server, mcp_server_auth_headers, mcp_auth_header
)
assert result == "Bearer zapier_token_upper"
def test_get_server_auth_header_with_server_name():
"""Test _get_server_auth_header function with server name (no alias)."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
_get_server_auth_header,
)
# Create a mock server with server_name but no alias
mock_server = MagicMock()
mock_server.alias = None
mock_server.server_name = "slack_server"
# Test with server-specific auth headers
mcp_server_auth_headers = {
"slack_server": "Bearer slack_token",
"zapier": "Bearer zapier_token",
}
mcp_auth_header = "Bearer default_token"
result = _get_server_auth_header(
mock_server, mcp_server_auth_headers, mcp_auth_header
)
assert result == "Bearer slack_token"
# Test case-insensitive matching
mcp_server_auth_headers = {
"SLACK_SERVER": "Bearer slack_token_upper",
"zapier": "Bearer zapier_token",
}
result = _get_server_auth_header(
mock_server, mcp_server_auth_headers, mcp_auth_header
)
assert result == "Bearer slack_token_upper"
def test_get_server_auth_header_fallback_to_default():
"""Test _get_server_auth_header function fallback to default auth header."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
_get_server_auth_header,
)
# Create a mock server
mock_server = MagicMock()
mock_server.alias = "unknown_server"
mock_server.server_name = "unknown_server_name"
# Test with no matching server-specific headers
mcp_server_auth_headers = {
"zapier": "Bearer zapier_token",
"slack": "Bearer slack_token",
}
mcp_auth_header = "Bearer default_token"
result = _get_server_auth_header(
mock_server, mcp_server_auth_headers, mcp_auth_header
)
assert result == "Bearer default_token"
# Test with no server-specific headers at all
result = _get_server_auth_header(mock_server, None, mcp_auth_header)
assert result == "Bearer default_token"
def test_get_server_auth_header_hyphenated_alias_sanitized_header_key():
"""Header keys use sanitized alias; lookup must match legacy hyphenated aliases."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
_get_server_auth_header,
)
mock_server = MagicMock()
mock_server.alias = "GitHub-MCP"
mock_server.server_name = "github_mcp_server"
mcp_server_auth_headers = {
"github_mcp": {"Authorization": "Bearer github-mcp-token"},
}
result = _get_server_auth_header(
mock_server, mcp_server_auth_headers, "Bearer default_token"
)
assert result == {"Authorization": "Bearer github-mcp-token"}
def test_get_server_auth_header_no_auth_headers():
"""Test _get_server_auth_header function with no auth headers."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
_get_server_auth_header,
)
# Create a mock server
mock_server = MagicMock()
mock_server.alias = "zapier"
mock_server.server_name = "zapier_server"
# Test with no auth headers
result = _get_server_auth_header(mock_server, None, None)
assert result is None
result = _get_server_auth_header(mock_server, {}, None)
assert result is None
def test_create_tool_response_objects():
"""Test _create_tool_response_objects enriches mcp_info with server_id and alias."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
_create_tool_response_objects,
)
from litellm.types.mcp_server.mcp_server_manager import MCPServer
from mcp.types import Tool as MCPTool
# Create mock tools
mock_tools = [
MCPTool(
name="send_email",
description="Send an email",
inputSchema={"type": "object", "properties": {"to": {"type": "string"}}},
),
MCPTool(
name="create_event",
description="Create a calendar event",
inputSchema={"type": "object", "properties": {"title": {"type": "string"}}},
),
]
server = MCPServer(
server_id="a1b2c3d4",
name="zapier_internal",
alias="zapier",
transport="http",
mcp_info={
"server_name": "zapier_internal",
"logo_url": "https://zapier.com/logo.png",
},
)
result = _create_tool_response_objects(mock_tools, server)
expected_mcp_info = {
"server_name": "zapier_internal",
"logo_url": "https://zapier.com/logo.png",
"server_id": "a1b2c3d4",
"alias": "zapier",
}
assert len(result) == 2
assert result[0].name == "send_email"
assert result[0].description == "Send an email"
assert result[0].mcp_info == expected_mcp_info
assert result[1].name == "create_event"
assert result[1].description == "Create a calendar event"
assert result[1].mcp_info == expected_mcp_info
@pytest.mark.asyncio
async def test_get_tools_for_single_server():
"""Test _get_tools_for_single_server function."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
_get_tools_for_single_server,
)
from mcp.types import Tool as MCPTool
# Create a mock server (pin allowlist fields; MagicMock auto-attrs are truthy)
mock_server = MagicMock()
mock_server.mcp_info = {"server_name": "zapier"}
mock_server.server_id = "zapier_id"
mock_server.alias = "zapier_alias"
mock_server.allowed_tools = None
mock_server.disallowed_tools = None
# Create mock tools
mock_tools = [
MCPTool(
name="send_email",
description="Send an email",
inputSchema={"type": "object", "properties": {"to": {"type": "string"}}},
)
]
# Mock the global_mcp_server_manager
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints.global_mcp_server_manager"
) as mock_manager:
mock_manager._get_tools_from_server = AsyncMock(return_value=mock_tools)
result = await _get_tools_for_single_server(mock_server, "Bearer test_token")
# Verify the manager was called with correct parameters
mock_manager._get_tools_from_server.assert_called_once_with(
server=mock_server,
mcp_auth_header="Bearer test_token",
extra_headers=None,
add_prefix=False,
raw_headers=None,
user_api_key_auth=None,
)
# Verify the result
assert len(result) == 1
assert result[0].name == "send_email"
assert result[0].mcp_info == {
"server_name": "zapier",
"server_id": "zapier_id",
"alias": "zapier_alias",
}
@pytest.mark.asyncio
async def test_get_tools_for_single_server_applies_disallowed_tools_without_allowlist():
"""REST listing must honor disallowed_tools even when no allowlist is set."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import (
_get_tools_for_single_server,
)
from mcp.types import Tool as MCPTool
mock_server = MagicMock()
mock_server.mcp_info = {"server_name": "zapier"}
mock_server.name = "zapier"
mock_server.server_id = "zapier"
mock_server.allowed_tools = None
mock_server.disallowed_tools = ["send_email"]
mock_tools = [
MCPTool(
name="send_email",
description="Send an email",
inputSchema={"type": "object"},
),
MCPTool(
name="read_email",
description="Read an email",
inputSchema={"type": "object"},
),
]
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints.global_mcp_server_manager"
) as mock_manager:
mock_manager._get_tools_from_server = AsyncMock(return_value=mock_tools)
result = await _get_tools_for_single_server(mock_server, "Bearer test_token")
assert [tool.name for tool in result] == ["read_email"]
@pytest.mark.asyncio
async def test_list_tool_rest_api_with_server_specific_auth():
"""Test list_tool_rest_api with server-specific auth headers."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import list_tool_rest_api
from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import (
MCPRequestHandler,
)
from litellm.proxy._types import UserAPIKeyAuth
# Create mock request with server-specific auth headers
mock_request = MagicMock()
mock_request.headers = {
"authorization": "Bearer user_token",
"x-mcp-zapier-authorization": "Bearer zapier_token",
"x-mcp-slack-authorization": "Bearer slack_token",
}
# Mock the MCPRequestHandler methods
with patch.object(
MCPRequestHandler, "_get_mcp_auth_header_from_headers"
) as mock_get_auth:
with patch.object(
MCPRequestHandler, "_get_mcp_server_auth_headers_from_headers"
) as mock_get_server_auth:
mock_get_auth.return_value = "Bearer default_token"
mock_get_server_auth.return_value = {
"zapier": "Bearer zapier_token",
"slack": "Bearer slack_token",
}
# Mock the global_mcp_server_manager
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints.global_mcp_server_manager"
) as mock_manager:
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["test-server-123"]
)
# Create a mock server
mock_server = MagicMock()
mock_server.server_id = "test-server-123"
mock_server.alias = "zapier"
mock_server.name = "zapier_server"
mock_server.mcp_info = {"server_name": "zapier"}
mock_manager.get_mcp_server_by_id.return_value = mock_server
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
mock_user_api_key_dict = UserAPIKeyAuth(
api_key="test",
user_id="test_user",
object_permission=LiteLLM_ObjectPermissionTable(
object_permission_id="dummy",
mcp_servers=[mock_server.server_id],
),
)
# Mock the _get_tools_for_single_server function
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints._get_tools_for_single_server"
) as mock_get_tools:
from litellm.proxy._experimental.mcp_server.server import (
ListMCPToolsRestAPIResponseObject,
)
mock_tools = [
ListMCPToolsRestAPIResponseObject(
name="send_email",
description="Send an email",
inputSchema={"type": "object"},
mcp_info={"server_name": "zapier"},
)
]
mock_get_tools.return_value = mock_tools
# Call the function
result = await list_tool_rest_api(
request=mock_request,
server_id="test-server-123",
user_api_key_dict=mock_user_api_key_dict,
)
# Verify the result
assert result["error"] is None
assert len(result["tools"]) == 1
assert result["tools"][0].name == "send_email"
# Verify that _get_tools_for_single_server was called with the correct auth header
mock_get_tools.assert_called_once()
call_args = mock_get_tools.call_args
assert call_args[0][0] == mock_server # server
assert (
call_args[0][1] == "Bearer zapier_token"
) # server_auth_header
@pytest.mark.asyncio
async def test_list_tool_rest_api_with_default_auth():
"""Test list_tool_rest_api with default auth header when no server-specific header is found."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import list_tool_rest_api
from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import (
MCPRequestHandler,
)
from litellm.proxy._types import UserAPIKeyAuth
# Create mock request with default auth header only
mock_request = MagicMock()
mock_request.headers = {
"authorization": "Bearer user_token",
"x-mcp-authorization": "Bearer default_token",
}
# Mock the MCPRequestHandler methods
with patch.object(
MCPRequestHandler, "_get_mcp_auth_header_from_headers"
) as mock_get_auth:
with patch.object(
MCPRequestHandler, "_get_mcp_server_auth_headers_from_headers"
) as mock_get_server_auth:
mock_get_auth.return_value = "Bearer default_token"
mock_get_server_auth.return_value = {} # No server-specific headers
# Mock the global_mcp_server_manager
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints.global_mcp_server_manager"
) as mock_manager:
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["test-server-123"]
)
# Create a mock server
mock_server = MagicMock()
mock_server.server_id = "test-server-123"
mock_server.alias = "unknown_server"
mock_server.name = "unknown_server"
mock_server.mcp_info = {"server_name": "unknown_server"}
mock_manager.get_mcp_server_by_id.return_value = mock_server
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
mock_user_api_key_dict = UserAPIKeyAuth(
api_key="test",
user_id="test_user",
object_permission=LiteLLM_ObjectPermissionTable(
object_permission_id="dummy",
mcp_servers=[mock_server.server_id],
),
)
# Mock the _get_tools_for_single_server function
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints._get_tools_for_single_server"
) as mock_get_tools:
from litellm.proxy._experimental.mcp_server.server import (
ListMCPToolsRestAPIResponseObject,
)
mock_tools = [
ListMCPToolsRestAPIResponseObject(
name="send_email",
description="Send an email",
inputSchema={"type": "object"},
mcp_info={"server_name": "unknown_server"},
)
]
mock_get_tools.return_value = mock_tools
# Call the function
result = await list_tool_rest_api(
request=mock_request,
server_id="test-server-123",
user_api_key_dict=mock_user_api_key_dict,
)
# Verify the result
assert result["error"] is None
assert len(result["tools"]) == 1
assert result["tools"][0].name == "send_email"
# Verify that _get_tools_for_single_server was called with the default auth header
mock_get_tools.assert_called_once()
call_args = mock_get_tools.call_args
assert call_args[0][0] == mock_server # server
assert (
call_args[0][1] == "Bearer default_token"
) # server_auth_header
@pytest.mark.asyncio
async def test_list_tool_rest_api_all_servers_with_auth():
"""Test list_tool_rest_api for all servers with server-specific auth headers."""
from litellm.proxy._experimental.mcp_server.rest_endpoints import list_tool_rest_api
from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import (
MCPRequestHandler,
)
from litellm.proxy._types import UserAPIKeyAuth
# Create mock request with server-specific auth headers
mock_request = MagicMock()
mock_request.headers = {
"authorization": "Bearer user_token",
"x-mcp-zapier-authorization": "Bearer zapier_token",
"x-mcp-slack-authorization": "Bearer slack_token",
}
# Mock the MCPRequestHandler methods
with patch.object(
MCPRequestHandler, "_get_mcp_auth_header_from_headers"
) as mock_get_auth:
with patch.object(
MCPRequestHandler, "_get_mcp_server_auth_headers_from_headers"
) as mock_get_server_auth:
mock_get_auth.return_value = "Bearer default_token"
mock_get_server_auth.return_value = {
"zapier": "Bearer zapier_token",
"slack": "Bearer slack_token",
}
# Mock the global_mcp_server_manager
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints.global_mcp_server_manager"
) as mock_manager:
# Create mock servers
mock_zapier_server = MagicMock()
mock_zapier_server.alias = "zapier"
mock_zapier_server.server_name = "zapier_server"
mock_zapier_server.mcp_info = {"server_name": "zapier"}
mock_slack_server = MagicMock()
mock_slack_server.alias = "slack"
mock_slack_server.server_name = "slack_server"
mock_slack_server.mcp_info = {"server_name": "slack"}
mock_manager.get_registry.return_value = {
"zapier": mock_zapier_server,
"slack": mock_slack_server,
}
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["zapier", "slack"]
)
mock_manager.get_mcp_server_by_id.side_effect = (
lambda server_id: mock_manager.get_registry.return_value.get(
server_id
)
)
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
mock_user_api_key_dict = UserAPIKeyAuth(
api_key="test",
user_id="test_user",
object_permission=LiteLLM_ObjectPermissionTable(
object_permission_id="dummy",
mcp_servers=["zapier", "slack"],
),
)
# Mock the _get_tools_for_single_server function
with patch(
"litellm.proxy._experimental.mcp_server.rest_endpoints._get_tools_for_single_server"
) as mock_get_tools:
from litellm.proxy._experimental.mcp_server.server import (
ListMCPToolsRestAPIResponseObject,
)
# Mock tools for each server
mock_get_tools.side_effect = [
[
ListMCPToolsRestAPIResponseObject(
name="send_email",
description="Send an email",
inputSchema={"type": "object"},
mcp_info={"server_name": "zapier"},
)
],
[
ListMCPToolsRestAPIResponseObject(
name="send_message",
description="Send a message",
inputSchema={"type": "object"},
mcp_info={"server_name": "slack"},
)
],
]
# Call the function without server_id (query all servers)
result = await list_tool_rest_api(
request=mock_request,
server_id=None,
user_api_key_dict=mock_user_api_key_dict,
)
# Verify the result
assert result["error"] is None
assert len(result["tools"]) == 2
assert result["tools"][0].name == "send_email"
assert result["tools"][1].name == "send_message"
# Verify that _get_tools_for_single_server was called for both servers
assert mock_get_tools.call_count == 2
server_auth_map = {
call_args[0][0]: call_args[0][1]
for call_args in mock_get_tools.call_args_list
}
assert (
server_auth_map.get(mock_zapier_server) == "Bearer zapier_token"
)
assert (
server_auth_map.get(mock_slack_server) == "Bearer slack_token"
)
@pytest.mark.asyncio
async def test_filter_tools_by_allowed_tools_integration():
"""Test that filter_tools_by_allowed_tools works correctly via _get_tools_from_mcp_servers"""
from litellm.proxy._experimental.mcp_server.server import (
_get_tools_from_mcp_servers,
)
from litellm.proxy._types import UserAPIKeyAuth
from mcp.types import Tool as MCPTool
# Create a mock user auth
mock_user_auth = UserAPIKeyAuth(api_key="test_key", user_id="test_user")
# Create mock tools that will be returned by the server
mock_tools = [
MCPTool(
name="allowed_tool_1",
description="This tool should be allowed",
inputSchema={"type": "object"},
),
MCPTool(
name="allowed_tool_2",
description="This tool should also be allowed",
inputSchema={"type": "object"},
),
MCPTool(
name="blocked_tool_1",
description="This tool should be blocked",
inputSchema={"type": "object"},
),
MCPTool(
name="blocked_tool_2",
description="This tool should also be blocked",
inputSchema={"type": "object"},
),
]
# Create a mock server with allowed_tools restriction
mock_server = MCPServer(
server_id="test-server-123",
name="test_server_with_allowed_tools",
url="https://test-server.com/mcp",
transport=MCPTransport.http,
allowed_tools=[
"allowed_tool_1",
"allowed_tool_2",
], # Only these tools should be returned
disallowed_tools=None,
)
# Create a mock MCPClient that returns all tools
mock_client = AsyncMock()
mock_client.list_tools = AsyncMock(return_value=mock_tools)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
def mock_client_constructor(*args, **kwargs):
return mock_client
# Mock the global MCP server manager
with patch(
"litellm.proxy._experimental.mcp_server.server.global_mcp_server_manager"
) as mock_manager:
# Mock manager methods
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["test-server-123"]
)
mock_manager.get_mcp_server_by_id = MagicMock(return_value=mock_server)
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
# Mock the _get_tools_from_server method to return all tools
mock_manager._get_tools_from_server = AsyncMock(return_value=mock_tools)
# Mock the MCPClient constructor
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Call _get_tools_from_mcp_servers which should apply the filtering
filtered_tools = await _get_tools_from_mcp_servers(
user_api_key_auth=mock_user_auth,
mcp_auth_header="Bearer test_token",
mcp_servers=None, # Get from all servers
)
# Verify that only allowed tools are returned
assert (
len(filtered_tools) == 2
), f"Expected 2 tools, got {len(filtered_tools)}"
tool_names = [tool.name for tool in filtered_tools]
assert (
"allowed_tool_1" in tool_names
), "allowed_tool_1 should be in filtered results"
assert (
"allowed_tool_2" in tool_names
), "allowed_tool_2 should be in filtered results"
assert (
"blocked_tool_1" not in tool_names
), "blocked_tool_1 should be filtered out"
assert (
"blocked_tool_2" not in tool_names
), "blocked_tool_2 should be filtered out"
# Verify the manager methods were called correctly
mock_manager.get_allowed_mcp_servers.assert_called_once_with(mock_user_auth)
# Note: get_mcp_server_by_id is now called for each server ID instead of batch
# Verify it was called with the correct server ID
assert mock_manager.get_mcp_server_by_id.call_count > 0
mock_manager._get_tools_from_server.assert_called_once()
@pytest.mark.asyncio
async def test_filter_tools_by_disallowed_tools_integration():
"""Test that filter_tools_by_allowed_tools works correctly with disallowed_tools via _get_tools_from_mcp_servers"""
from litellm.proxy._experimental.mcp_server.server import (
_get_tools_from_mcp_servers,
)
from litellm.proxy._types import UserAPIKeyAuth
from mcp.types import Tool as MCPTool
# Create a mock user auth
mock_user_auth = UserAPIKeyAuth(api_key="test_key", user_id="test_user")
# Create mock tools that will be returned by the server
mock_tools = [
MCPTool(
name="safe_tool_1",
description="This tool should be allowed",
inputSchema={"type": "object"},
),
MCPTool(
name="safe_tool_2",
description="This tool should also be allowed",
inputSchema={"type": "object"},
),
MCPTool(
name="dangerous_tool_1",
description="This tool should be blocked",
inputSchema={"type": "object"},
),
MCPTool(
name="dangerous_tool_2",
description="This tool should also be blocked",
inputSchema={"type": "object"},
),
]
# Create a mock server with disallowed_tools restriction
mock_server = MCPServer(
server_id="test-server-456",
name="test_server_with_disallowed_tools",
url="https://test-server.com/mcp",
transport=MCPTransport.http,
allowed_tools=None,
disallowed_tools=[
"dangerous_tool_1",
"dangerous_tool_2",
], # These tools should be filtered out
)
# Create a mock MCPClient that returns all tools
mock_client = AsyncMock()
mock_client.list_tools = AsyncMock(return_value=mock_tools)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
def mock_client_constructor(*args, **kwargs):
return mock_client
# Mock the global MCP server manager
with patch(
"litellm.proxy._experimental.mcp_server.server.global_mcp_server_manager"
) as mock_manager:
# Mock manager methods
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["test-server-456"]
)
mock_manager.get_mcp_server_by_id = MagicMock(return_value=mock_server)
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
# Mock the _get_tools_from_server method to return all tools
mock_manager._get_tools_from_server = AsyncMock(return_value=mock_tools)
# Mock the MCPClient constructor
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Call _get_tools_from_mcp_servers which should apply the filtering
filtered_tools = await _get_tools_from_mcp_servers(
user_api_key_auth=mock_user_auth,
mcp_auth_header="Bearer test_token",
mcp_servers=None, # Get from all servers
)
# Verify that only safe tools are returned (dangerous tools filtered out)
assert (
len(filtered_tools) == 2
), f"Expected 2 tools, got {len(filtered_tools)}"
tool_names = [tool.name for tool in filtered_tools]
assert (
"safe_tool_1" in tool_names
), "safe_tool_1 should be in filtered results"
assert (
"safe_tool_2" in tool_names
), "safe_tool_2 should be in filtered results"
assert (
"dangerous_tool_1" not in tool_names
), "dangerous_tool_1 should be filtered out"
assert (
"dangerous_tool_2" not in tool_names
), "dangerous_tool_2 should be filtered out"
# Verify the manager methods were called correctly
mock_manager.get_allowed_mcp_servers.assert_called_once_with(mock_user_auth)
# Note: get_mcp_server_by_id is now called for each server ID instead of batch
# Verify it was called with the correct server ID
assert mock_manager.get_mcp_server_by_id.call_count > 0
mock_manager._get_tools_from_server.assert_called_once()
@pytest.mark.asyncio
async def test_filter_tools_no_restrictions_integration():
"""Test that filter_tools_by_allowed_tools returns all tools when no restrictions are set"""
from litellm.proxy._experimental.mcp_server.server import (
_get_tools_from_mcp_servers,
)
from litellm.proxy._types import UserAPIKeyAuth
from mcp.types import Tool as MCPTool
# Create a mock user auth
mock_user_auth = UserAPIKeyAuth(api_key="test_key", user_id="test_user")
# Create mock tools that will be returned by the server
mock_tools = [
MCPTool(
name="tool_1",
description="Tool 1",
inputSchema={"type": "object"},
),
MCPTool(
name="tool_2",
description="Tool 2",
inputSchema={"type": "object"},
),
]
# Create a mock server with no tool restrictions
mock_server = MCPServer(
server_id="test-server-000",
name="test_server_no_restrictions",
url="https://test-server.com/mcp",
transport=MCPTransport.http,
allowed_tools=None, # No restrictions
disallowed_tools=None, # No restrictions
)
# Create a mock MCPClient that returns all tools
mock_client = AsyncMock()
mock_client.list_tools = AsyncMock(return_value=mock_tools)
mock_client.__aenter__ = AsyncMock(return_value=mock_client)
mock_client.__aexit__ = AsyncMock(return_value=None)
def mock_client_constructor(*args, **kwargs):
return mock_client
# Mock the global MCP server manager
with patch(
"litellm.proxy._experimental.mcp_server.server.global_mcp_server_manager"
) as mock_manager:
# Mock manager methods
mock_manager.get_allowed_mcp_servers = AsyncMock(
return_value=["test-server-000"]
)
mock_manager.get_mcp_server_by_id = MagicMock(return_value=mock_server)
# Mock filter_server_ids_by_ip_with_info to return input unchanged (no IP filtering in test)
mock_manager.filter_server_ids_by_ip_with_info = MagicMock(
side_effect=lambda server_ids, client_ip: (server_ids, 0)
)
# Mock the _get_tools_from_server method to return all tools
mock_manager._get_tools_from_server = AsyncMock(return_value=mock_tools)
# Mock the MCPClient constructor
with patch(
"litellm.proxy._experimental.mcp_server.mcp_server_manager.MCPClient",
mock_client_constructor,
):
# Call _get_tools_from_mcp_servers which should apply the filtering
filtered_tools = await _get_tools_from_mcp_servers(
user_api_key_auth=mock_user_auth,
mcp_auth_header="Bearer test_token",
mcp_servers=None, # Get from all servers
)
# Should return all tools when no restrictions
assert (
len(filtered_tools) == 2
), f"Expected 2 tools, got {len(filtered_tools)}"
tool_names = [tool.name for tool in filtered_tools]
assert "tool_1" in tool_names, "tool_1 should be in filtered results"
assert "tool_2" in tool_names, "tool_2 should be in filtered results"
@pytest.mark.asyncio
async def test_mcp_access_group_permission_inheritance_integration():
"""Integration test for MCP access group permission inheritance"""
from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import (
MCPRequestHandler,
)
from litellm.proxy._types import UserAPIKeyAuth
# Test scenario: team has access groups, key has no permissions -> should inherit
# Use direct mocking of the helper functions instead of complex database mocking
with patch.object(
MCPRequestHandler, "_get_allowed_mcp_servers_for_key"
) as mock_key:
with patch.object(
MCPRequestHandler, "_get_allowed_mcp_servers_for_team"
) as mock_team:
# Key has no permissions, team has servers
mock_key.return_value = [] # Key inherits nothing directly
mock_team.return_value = [
"staff-server-1",
"staff-server-2",
"ops-server-1",
] # Team has servers
# Create user auth object
user_auth = UserAPIKeyAuth(
api_key="test-key",
user_id="test-user",
team_id="team-staff",
object_permission_id=None, # Key has no explicit permissions
)
# Test the inheritance logic
allowed_servers = await MCPRequestHandler.get_allowed_mcp_servers(user_auth)
# Should inherit all team servers since key has no permissions
expected_servers = ["staff-server-1", "staff-server-2", "ops-server-1"]
assert sorted(allowed_servers) == sorted(expected_servers)
@pytest.mark.asyncio
async def test_mcp_access_group_permission_intersection_integration():
"""Integration test for MCP access group permission intersection"""
from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import (
MCPRequestHandler,
)
from litellm.proxy._types import UserAPIKeyAuth
# Test scenario: both team and key have access groups -> should intersect
# Use direct mocking of the helper functions instead of complex database mocking
with patch.object(
MCPRequestHandler, "_get_allowed_mcp_servers_for_key"
) as mock_key:
with patch.object(
MCPRequestHandler, "_get_allowed_mcp_servers_for_team"
) as mock_team:
# Both key and team have permissions - should intersect
mock_key.return_value = [
"ops-server",
"external-server",
] # Key has these servers
mock_team.return_value = [
"staff-server",
"ops-server",
"admin-server",
] # Team has these servers
# Create user auth object
user_auth = UserAPIKeyAuth(
api_key="test-key",
user_id="test-user",
team_id="team-staff",
object_permission_id="key-permission-id", # Key has explicit permissions
)
# Test the intersection logic
allowed_servers = await MCPRequestHandler.get_allowed_mcp_servers(user_auth)
# Should only get intersection (ops-server is common)
expected_servers = ["ops-server"]
assert sorted(allowed_servers) == sorted(expected_servers)
@pytest.mark.asyncio
async def test_mcp_server_manager_with_access_groups_integration():
"""Integration test for MCPServerManager with access group filtering"""
from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import (
MCPRequestHandler,
)
from litellm.proxy._types import UserAPIKeyAuth
# Create a test manager
test_manager = MCPServerManager()
# Load servers with access groups
await test_manager.load_servers_from_config(
{
"staff_server": {
"url": "https://staff-server.com/mcp",
"access_groups": ["staff"],
"transport": MCPTransport.http,
},
"ops_server": {
"url": "https://ops-server.com/mcp",
"access_groups": ["ops"],
"transport": MCPTransport.http,
},
"admin_server": {
"url": "https://admin-server.com/mcp",
"access_groups": ["admin"],
"transport": MCPTransport.http,
},
}
)
# Mock user with specific access groups
user_auth = UserAPIKeyAuth(
api_key="test-key", user_id="test-user", team_id="team-staff"
)
# Mock the permission lookup to return staff access group
with patch.object(MCPRequestHandler, "get_allowed_mcp_servers") as mock_get_allowed:
mock_get_allowed.return_value = [
"staff-server-id",
"ops-server-id",
] # User has access to staff and ops
allowed_servers = await test_manager.get_allowed_mcp_servers(user_auth)
# Should only get servers user has access to
assert len(allowed_servers) >= 0 # At least verify no errors
mock_get_allowed.assert_called_once_with(user_auth)
@pytest.mark.asyncio
async def test_get_allowed_mcp_servers_returns_registry_for_admin():
from litellm.proxy._types import LitellmUserRoles, UserAPIKeyAuth
from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import (
MCPRequestHandler,
)
test_manager = MCPServerManager()
await test_manager.load_servers_from_config(
{
"alpha_server": {
"url": "https://alpha.server/mcp",
"transport": MCPTransport.http,
},
"beta_server": {
"url": "https://beta.server/mcp",
"transport": MCPTransport.http,
},
}
)
admin_auth = UserAPIKeyAuth(
api_key="admin-key",
user_role=LitellmUserRoles.PROXY_ADMIN,
)
with patch.object(
MCPRequestHandler, "get_allowed_mcp_servers", new_callable=AsyncMock
) as mock_permission_lookup:
allowed_servers = await test_manager.get_allowed_mcp_servers(admin_auth)
assert set(allowed_servers) == set(test_manager.get_registry().keys())
mock_permission_lookup.assert_not_called()
@pytest.mark.asyncio
async def test_get_allowed_mcp_servers_returns_empty_for_non_admin_without_permissions():
from litellm.proxy._types import LitellmUserRoles, UserAPIKeyAuth
from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import (
MCPRequestHandler,
)
test_manager = MCPServerManager()
await test_manager.load_servers_from_config(
{
"alpha_server": {
"url": "https://alpha.server/mcp",
"transport": MCPTransport.http,
},
"beta_server": {
"url": "https://beta.server/mcp",
"transport": MCPTransport.http,
},
}
)
user_auth = UserAPIKeyAuth(
api_key="user-key",
user_role=LitellmUserRoles.INTERNAL_USER,
)
with patch.object(
MCPRequestHandler, "get_allowed_mcp_servers", new_callable=AsyncMock
) as mock_permission_lookup:
mock_permission_lookup.return_value = []
allowed_servers = await test_manager.get_allowed_mcp_servers(user_auth)
assert allowed_servers == []
mock_permission_lookup.assert_awaited_once()
@pytest.mark.asyncio
async def test_call_mcp_tool_uses_manager_permission_lookup():
from litellm.proxy._experimental.mcp_server.server import (
call_mcp_tool,
global_mcp_server_manager,
)
mock_server = MCPServer(
server_id="server-123",
name="test_server",
alias="test_server",
server_name="test_server",
url="https://test-server.com/mcp",
transport=MCPTransport.http,
mcp_info={"server_name": "test_server"},
)
expected_response = [TextContent(type="text", text="ok")]
with (
patch.object(
global_mcp_server_manager,
"get_allowed_mcp_servers",
new_callable=AsyncMock,
) as mock_get_allowed,
patch.object(
global_mcp_server_manager,
"get_mcp_server_by_id",
return_value=mock_server,
),
patch.object(
global_mcp_server_manager,
"_get_mcp_server_from_tool_name",
return_value=mock_server,
) as mock_get_server,
patch(
"litellm.proxy._experimental.mcp_server.server.global_mcp_tool_registry"
) as mock_tool_registry,
patch(
"litellm.proxy._experimental.mcp_server.server._handle_managed_mcp_tool",
new_callable=AsyncMock,
) as mock_handle_managed,
patch(
"litellm.proxy._experimental.mcp_server.server.MCPRequestHandler.is_tool_allowed",
return_value=True,
),
):
mock_get_allowed.return_value = [mock_server.server_id]
mock_tool_registry.get_tool.return_value = None
mock_handle_managed.return_value = expected_response
result = await call_mcp_tool(
name=f"{mock_server.name}/gmail_send_email",
arguments={"body": "hello"},
mcp_servers=["test_server"],
)
assert result == expected_response
mock_get_allowed.assert_awaited_once()
# We call `_get_mcp_server_from_tool_name` multiple times:
# - for logging/metadata
# - for resolving the server during dispatch
# - and inside `call_tool` for guardrails/hooks
# The exact count isn't important, only that it is used.
assert mock_get_server.call_count >= 2
# First call should use the prefixed tool name
assert (
mock_get_server.call_args_list[0][0][0]
== f"{mock_server.name}/gmail_send_email"
)
@pytest.mark.asyncio
async def test_call_mcp_tool_resolves_unprefixed_tool_name_and_checks_permissions():
"""
Ensure `call_mcp_tool` correctly resolves the MCP server for an unprefixed tool
name and enforces server-level permissions using that resolved server.
"""
from litellm.proxy._experimental.mcp_server.server import (
call_mcp_tool,
global_mcp_server_manager,
)
mock_server = MCPServer(
server_id="server-123",
name="test_server",
alias="test_server",
server_name="test_server",
url="https://test-server.com/mcp",
transport=MCPTransport.http,
mcp_info={"server_name": "test_server"},
)
expected_response = [TextContent(type="text", text="ok")]
with (
patch.object(
global_mcp_server_manager,
"get_allowed_mcp_servers",
new_callable=AsyncMock,
) as mock_get_allowed,
patch.object(
global_mcp_server_manager,
"get_mcp_server_by_id",
return_value=mock_server,
),
patch.object(
global_mcp_server_manager,
"_get_mcp_server_from_tool_name",
return_value=mock_server,
) as mock_get_server,
patch(
"litellm.proxy._experimental.mcp_server.server.global_mcp_tool_registry"
) as mock_tool_registry,
patch(
"litellm.proxy._experimental.mcp_server.server._handle_managed_mcp_tool",
new_callable=AsyncMock,
) as mock_handle_managed,
patch(
"litellm.proxy._experimental.mcp_server.server.MCPRequestHandler.is_tool_allowed",
return_value=True,
) as mock_is_allowed,
):
mock_get_allowed.return_value = [mock_server.server_id]
mock_tool_registry.get_tool.return_value = None
mock_handle_managed.return_value = expected_response
# Call with UNPREFIXED tool name; server should be resolved via mapping
result = await call_mcp_tool(
name="gmail_send_email",
arguments={"body": "hello"},
mcp_servers=["test_server"],
)
assert result == expected_response
mock_get_allowed.assert_awaited_once()
# We should resolve the server at least once using the unprefixed name
assert mock_get_server.call_count >= 1
assert mock_get_server.call_args_list[0][0][0] == "gmail_send_email"
# Permissions check should be invoked with the resolved server name
mock_is_allowed.assert_called_once()