Fix compress_schema to preserve additionalProperties: false for MCP compatibility

[jlowin]

Fixes #3008

Summary

The compress_schema function was defaulting to prune_additional_properties=True, which removed additionalProperties: false from JSON schemas. This broke compatibility with MCP clients like Claude that require strict JSON schemas with additionalProperties: false.

Changes

• Changed default of prune_additional_properties from True to False in compress_schema
• Added test demonstrating MCP client compatibility requirement
• Updated existing tests to explicitly enable pruning when needed
• Added additionalProperties: false to manually constructed schemas in tool_transform.py
• Updated inline snapshots to reflect new behavior

Testing

• All 3688 tests pass
• New test validates the fix
• Linting and type checking pass

Generated with Claude Code) | View branch | [View job run](https://github.com/jlowin/fastmcp/actions/runs/21443654395

[coderabbitai]

Walkthrough

The pull request modifies JSON schema handling across two files to ensure stricter validation of tool parameters. In src/fastmcp/tools/tool_transform.py, two schema generation functions now explicitly include "additionalProperties": False in their output. Additionally, src/fastmcp/utilities/json_schema.py changes the default value of the prune_additional_properties parameter in the compress_schema function from True to False, preserving the additionalProperties: false constraint during schema compression rather than removing it.

Possibly related PRs

• Dereference $ref in tool schemas for MCP client compatibility #2808: Modifies JSON schema processing behavior including compress_schema functionality and its downstream usage in tool schema generation, directly overlapping with the schema strictness changes in this PR.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and accurately describes the main change: fixing compress_schema to preserve additionalProperties: false for MCP compatibility.
Description check	✅ Passed	The description includes a clear summary, detailed changes, and testing results. The required checklist items are present, though some may not be explicitly checked off.
Linked Issues check	✅ Passed	The PR addresses all requirements from issue #3008: changed default of prune_additional_properties from True to False, added tests for MCP compatibility, and updated tests to explicitly enable pruning when needed.
Out of Scope Changes check	✅ Passed	All changes are directly related to fixing issue #3008: modifications to compress_schema default parameter, tool_transform.py schema construction, and related test updates are within scope.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch claude/issue-3008-20260128-1511

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[marvin-context-protocol]

Test Failure Analysis

Summary: The Windows CI job is timing out in tests/cli/test_run.py::TestMCPConfig::test_run_mcp_config after exceeding the default 5-second pytest timeout.

Root Cause: The test creates a subprocess-based MCP server and connects to it via a Client. Windows subprocess creation and stdio communication is significantly slower than Linux, causing the test to exceed the 5-second timeout. The test passes on all Linux runners (Ubuntu with Python 3.10, 3.13, and lowest-direct dependencies) but consistently times out on Windows.

Suggested Solution: Add a @pytest.mark.timeout(10) decorator to the test_run_mcp_config test method in tests/cli/test_run.py at line 99. This follows the established pattern used in other subprocess/transport tests:

• tests/client/transports/test_uv_transport.py: Uses 10-second timeout for subprocess transport tests
• tests/integration_tests/test_timeout_fix.py: Uses 15-second timeout for integration tests
• tests/server/providers/openapi/test_openapi_performance.py: Uses 10-second timeout for performance tests

Detailed Analysis

Test Execution Log

The test begins at line 430 and times out after the 5-second default:

tests\cli\test_run.py ........+++++++++++++++++++++++++++++++++++ Timeout +++++++++++++++++++++++++++++++++++
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Captured stderr ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
                    DEBUG    Inferred transport:               inference.py:153
                             <MCPConfigTransport(config='mcpServers={'test_server':
                             StdioMCPServer(command='python',
                             args=['C:\\Users\\runneradmin\\AppData\\Local\\Temp\\pytest-of-runneradmin\\pytest-0\\test_run_mcp_config0\\test.py']
...
                    DEBUG    Stdio transport connected             stdio.py:191

The last log line shows "Stdio transport connected", indicating the subprocess connection was established but the test timed out before completing the client.list_tools() call at line 131.

Why This Fails on Windows Only

• Windows subprocess creation is 2-4x slower than Linux due to process model differences
• Windows stdio buffering behaves differently, potentially causing additional delays
• The test involves: subprocess spawn → stdio connection → MCP handshake → list_tools RPC call
• On Linux this completes in ~2-3 seconds; on Windows it takes ~5-7 seconds

Related Files

• tests/cli/test_run.py:99 - The failing test that needs the timeout marker
• tests/client/transports/test_uv_transport.py:20,57 - Similar subprocess tests with 10s timeout
• pyproject.toml:125 - Global pytest timeout configuration (5 seconds)

---

🤖 This analysis was generated by marvin