tooltest-fix-loop
Use when running tooltest to validate MCP servers, interpret failures, and iterate fixes in this repo.
When & Why to Use This Skill
The tooltest-fix-loop skill is a specialized developer utility designed to automate the validation and debugging of Model Context Protocol (MCP) servers. It streamlines the end-to-end process of running diagnostic tests, interpreting complex failure reports across protocol and schema layers, and implementing iterative fixes to ensure robust server performance and specification compliance.
Use Cases
- Automated MCP Server Validation: Efficiently running tooltest via stdio or HTTP commands to verify the integrity of MCP server implementations.
- Protocol and Schema Debugging: Identifying and resolving failures in protocol handling, tool schema validation, or runtime behavior through systematic triage.
- Regression Testing and Maintenance: Ensuring backward compatibility and stability of tool names and schemas during server updates by maintaining a continuous test-fix loop.
- Failure Interpretation: Translating technical failure reports into actionable code patches to achieve a zero-exit status in testing environments.
| name | tooltest-fix-loop |
|---|---|
| description | Use when running tooltest to validate MCP servers, interpret failures, and iterate fixes in this repo. |
Tooltest fix loop
Run tooltest and iterate
- Identify how to start the repo’s MCP server (stdio or streamable HTTP).
- Run tooltest against it:
tooltest stdio --command "<server start command>"tooltest http --url "<server mcp url>"
- Read the failure report, trace the failing tool call or protocol step, and fix the smallest viable patch.
- Re-run tooltest until it exits 0; keep tool names and schemas stable unless a spec violation requires change.
Triage guidance
- Treat failures as either protocol handling, tool schema validation, or runtime behavior; fix in that order.
- Prefer backward-compatible fixes; add a regression test when a failure is deterministic or easy to reproduce.
- When the failure is intermittent, keep the smallest reproduction from the report and turn it into a test.
Reporting
- Summarize the changes and why they address the specific tooltest failures.
- Include the final passing tooltest output snippet.