fix(mcp): stop logging tool-call input in MCP client

[ryan-crabbe-berri]

Relevant issues

Linear ticket

LIT-3811

Pre-Submission checklist

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

• I have added meaningful tests
• 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

CI (LiteLLM team)

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Screenshots / Proof of Fix

The MCP client logged the tool input at INFO on every call, so the caller's arguments (the user query, model, instructions, etc.) showed up verbatim in the proxy logs and any downstream log aggregator. That is the line in the LIT-3811 screenshots

Proof against a live proxy running this branch's code, calling the public DeepWiki MCP server at production INFO level (no debug flags). Config used

mcp_servers:
  deepwiki:
    transport: "http"
    url: "https://mcp.deepwiki.com/mcp"
    allow_all_keys: true

Same real tool call in both runs, with a sentinel string in the question so it is easy to grep

curl -s http://localhost:4001/mcp-rest/tools/call \
  -H 'Authorization: Bearer sk-1234' -H 'Content-Type: application/json' \
  -d '{"server_id":"<deepwiki>","name":"ask_question","arguments":{"repoName":"BerriAI/litellm","question":"LIT3811SECRETPROBE what is litellm"}}'

Before the fix the INFO line carries the full arguments, and the input is grep-able in the logs

$ grep "MCP client calling tool" litellm.log
17:21:54 - LiteLLM:INFO: client.py:565 - MCP client calling tool 'ask_question' with arguments: {'repoName': 'BerriAI/litellm', 'question': 'LIT3811SECRETPROBE what is litellm'}

$ grep -c "LIT3811SECRETPROBE" litellm.log
1

After the fix the same line logs only the tool name, and the input no longer appears anywhere in the logs

$ grep "MCP client calling tool" litellm.log
17:22:34 - LiteLLM:INFO: client.py:565 - MCP client calling tool 'ask_question'

$ grep -c "LIT3811SECRETPROBE" litellm.log
0

Type

Bug Fix

Changes

The MCP client emitted the full tool arguments at INFO in call_tool, and the same for prompt arguments in get_prompt, so caller input flowed into the application logs on every call. Both lines now log only the tool or prompt name and drop the arguments. The name is a stable identifier and is not sensitive; the input is what must not be logged

Files: litellm/experimental_mcp_client/client.py (the two INFO log lines), plus regression tests in tests/test_litellm/experimental_mcp_client/test_mcp_client.py that drive call_tool / get_prompt with a sensitive arguments value and assert it never reaches the logger (they fail on the pre-fix code and pass on the fix)

[greptile-apps]

Greptile Summary

This PR removes caller-provided MCP arguments from client logs. The main changes are:

• Drops tool-call arguments from the call_tool INFO log
• Drops prompt arguments from the get_prompt INFO log
• Adds tests that verify sensitive argument values are not emitted while names still appear

Confidence Score: 5/5

The change is narrowly scoped to redacting sensitive MCP argument values from INFO logs while preserving useful operation names.

The implementation matches the stated privacy goal and includes regression coverage for both tool and prompt argument logging paths.

T-Rex Logs

What T-Rex did

• Ran the verbose logging verification for the MCP client tool call, capturing before and after logs that show the input arguments (hello-from-LIT-3811-call-tool and gpt-5-mini) were present before and removed from the verbose logger after.
• Compared the mcp-get-prompt-logging artifacts, confirming that the before log included the argument hello-from-LIT-3811-get-prompt and the after log omits it, while the get_prompt INFO log still shows my_prompt and the session call data remains visible.

_{Ran code and verified through T-Rex}

_{Reviews (2): Last reviewed commit: "fix(mcp): stop logging tool-call input i..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[ryan-crabbe-berri]

@greptileai re review