PR #24986: fix(mcp): preserve non-MCP tools in semantic filter

• Repository: BerriAI/litellm
• Author: lazyskyline7
• State: open | merged: False
• Link: https://github.com/BerriAI/litellm/pull/24986

Description (problem / solution / changelog)

Relevant issues

Fixes #24984

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/test_litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🐛 Bug Fix

Changes

The semantic filter hook treated all tools equally — it didn't distinguish between MCP tools (which it should filter) and non-MCP built-in tools (which should pass through untouched). This caused non-MCP tools to be dropped on match, and all tools returned on zero match.

• Move MCP/non-MCP separation into the hook (hook.py): only pass MCP tools (identified by server-name prefix via is_tool_name_prefixed()) to filter_tools(), always recombine with non-MCP tools after
• filter_tools() returns empty list on zero matches instead of all tools — the hook adds non-MCP tools back
• 2 new tests: zero-match returns empty, hook preserves non-MCP tools on match
• 1 existing test updated: tool names prefixed to reflect MCP tool format

Changed files

• litellm/proxy/_experimental/mcp_server/semantic_tool_filter.py (modified, +4/-1)
• litellm/proxy/hooks/mcp_semantic_filter/hook.py (modified, +22/-3)
• tests/test_litellm/proxy/_experimental/mcp_server/test_semantic_tool_filter.py (modified, +166/-8)

PR #24990: fix(mcp): return only non-MCP tools when semantic filter finds zero matches

• Repository: BerriAI/litellm
• Author: Vitalcheffe
• State: open | merged: False
• Link: https://github.com/BerriAI/litellm/pull/24990

Description (problem / solution / changelog)

What

When the semantic tool filter finds zero matches for a query, it now returns only non-MCP (built-in) tools instead of returning all available tools.

Why

When zero semantic matches exist (e.g., an irrelevant query like "hello"), filter_tools() fell back to returning ALL tools. With 100+ MCP tools configured, this exceeds OpenAI"s 128 tool limit and causes requests to fail. See: #24984

How

In semantic_tool_filter.py, the fallback when matched_tool_names is empty now filters to only non-MCP tools using is_tool_name_prefixed() from utils.py. MCP tools are identifiable by their server-name prefix (contains MCP_TOOL_PREFIX_SEPARATOR). Built-in tools like web_search or code_interpreter pass through unaffected.

Testing

• Added test_semantic_filter_zero_matches_returns_only_non_mcp_tools — verifies that on zero matches, only non-prefixed tools are returned
• Existing tests continue to pass
• ruff check passes

Checklist

• Tests added
• Linter passes
• No breaking change for non-MCP tools

Closes #24984

Changed files

• litellm/proxy/_experimental/mcp_server/semantic_tool_filter.py (modified, +15/-1)
• tests/test_litellm/proxy/_experimental/mcp_server/test_semantic_tool_filter.py (modified, +102/-0)