fix(mcp): update test mocks for renamed filter_server_ids_by_ip_with_info

[jquinter]

Summary

• MCP tests in test_mcp_server.py were mocking the old method name filter_server_ids_by_ip but production code (server.py:774) calls filter_server_ids_by_ip_with_info which returns a (server_ids, blocked_count) tuple
• The unmocked method on AsyncMock returned a coroutine instead of a tuple, causing TypeError: cannot unpack non-iterable coroutine object and ValueError: not enough values to unpack
• Updated all 11 mock definitions to use the correct method name and return the expected (server_ids, 0) tuple

Fixes 3 failing tests on main:

• test_get_tools_from_mcp_servers
• test_filter_tools_by_disallowed_tools_integration
• test_filter_tools_by_allowed_tools_integration

Test plan

• All 3 previously failing tests now pass locally
• CI MCP test suite passes

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Building	Preview, Comment	Feb 28, 2026 1:37am

[greptile-apps]

Greptile Summary

This PR fixes 3 failing MCP tests by updating mock definitions to match a renamed production method. The method filter_server_ids_by_ip was renamed to filter_server_ids_by_ip_with_info in mcp_server_manager.py, changing the return type from List[str] to Tuple[List[str], int]. The test file's 11 mock definitions were still using the old name and return format, causing TypeError and ValueError when the production code tried to unpack the tuple.

• All 11 mock definitions updated from filter_server_ids_by_ip to filter_server_ids_by_ip_with_info
• Mock return values updated from server_ids to (server_ids, 0) to match the new (filtered_ids, ip_blocked_count) tuple signature
• No remaining references to the old method name in the test file
• Changes are consistent with mocks in the newer test file at tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_server.py

Confidence Score: 5/5

• This PR is safe to merge — it only updates test mocks to match the already-renamed production method, with no production code changes.
• The change is a mechanical find-and-replace of mock method names and return values in a single test file. The new mock signatures exactly match the production code's method signature. No production code is modified.
• No files require special attention.

Important Files Changed

Filename	Overview
tests/mcp_tests/test_mcp_server.py	Updates 11 mock definitions from old method name filter_server_ids_by_ip to renamed filter_server_ids_by_ip_with_info, with correct (server_ids, 0) tuple return value matching production code signature.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["Test calls production code<br/>(server.py:777)"] --> B["global_mcp_server_manager<br/>.filter_server_ids_by_ip_with_info()"]
    B --> C{"Mock present?"}
    C -->|"Before fix: mocked as<br/>filter_server_ids_by_ip"| D["❌ Unmocked AsyncMock<br/>returns coroutine"]
    D --> E["TypeError: cannot unpack<br/>non-iterable coroutine object"]
    C -->|"After fix: mocked as<br/>filter_server_ids_by_ip_with_info"| F["✅ Returns (server_ids, 0)<br/>tuple as expected"]
    F --> G["Production code unpacks:<br/>ids, blocked = result"]
    G --> H["Test passes ✅"]

Loading

_{Last reviewed commit: cb4cfa1}

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[jquinter]

The Ruff PLR0915 lint failures are not related to this PR. They are pre-existing on main. Fix PR: #22328

[jquinter]

The e2e test failure (test_independent_clients_no_shared_session) is not related to this PR. It fails on main too — the test gets "Error: User not allowed to call this tool." instead of "30", which is an MCP authorization issue in the e2e test setup, not a mock naming issue.

[jquinter]

The test_independent_clients_no_shared_session e2e failure is pre-existing on main (all 4 MCP e2e tests fail locally). Root cause: set_auth_context() uses a ContextVar set during the HTTP handler, but the MCP SessionManager spawns separate anyio.TaskGroup tasks for protocol messages like call_tool. The ContextVar doesn't propagate into these new tasks, so get_auth_context() returns empty/stale data → "User not allowed to call this tool". This needs a separate issue/fix for auth context propagation in the MCP session manager.

[jquinter]

The realtime streaming guardrail test failures (test_realtime_guardrail_blocks_prompt_injection, test_realtime_text_input_guardrail_blocks_and_returns_error) are also pre-existing on main — test assertions didn't account for the "voice the guardrail message" feature. Fix PR: #22332

[jquinter]

CI Failure: Anthropic pass-through test + Bedrock parallel_tool_calls

Both failures are pre-existing on main and unrelated to this PR's MCP test mock changes.

1. Bedrock test_parallel_tool_calls_newer_model_adds_disable_flag

Root cause: Commit 8565c70e53 (revert) removed parallel_tool_calls from map_openai_params, and the subsequent re-fix d0445e1e33 only restored the transform_request consumer but not the producer.
Fix PR: #22333

2. Anthropic test_anthropic_experimental_pass_through_messages_handler_dynamic_api_key_and_api_base_and_custom_values

Root cause: Commit 99c62ca40e removed "azure" from _RESPONSES_API_PROVIDERS, routing Azure models through litellm.completion instead of litellm.responses. The test was not updated to match.
Fix PR: #22334

[jquinter]

CI Failure: Schema migration check

The test_aaaasschema_migration_check failure is pre-existing on main and unrelated to this PR.

Root cause: PR #22271 added LiteLLM_ClaudeCodePluginTable to schema.prisma but did not include a corresponding migration file.

Fix PR: #22335

[jquinter]

@shin-bot-litellm evaluate this PR