langfuse-mcp-general-rules

How to build, run, test and debug Langfuse MCP project effectively

• We ONLY use uv/uvx in order to run, build our app, do not use vanilla Python or scripts
• Our server logs reside in /tmp/langfuse_mcp.log - read them in our for properly debugging our server
• For debugging Cursor MCP client, the logs exist in: ~/Library/Application\ Support/Cursor/logs/
• In order to only find MCP related logs one can use find ~/Library/Application\ Support/Cursor/logs/ -type f -exec grep -i "mcp" {} \;
• Since we aren't using SSE and there are many SSE implementations that are used, one can also use find ~/Library/Application\ Support/Cursor/logs/ -type f -exec grep -i "mcp" {} \; | grep -v "SSE error"
• Read README.md for various information, which relates to our project.

Building and Publishing to PyPI

Version Numbering and Tagging

• We use semantic versioning with format v0.x.y for version tags
• Always create a git tag for each release with the format v0.x.y

Build and Publish Workflow

1. Commit your changes:

   git add <modified files>
   git commit -m "Your commit message"
   

2. Tag the release:

   git tag -a v0.x.y -m "Brief description of the release"
   

3. Build the wheel (skip source distribution):

   rm -rf dist/
   uv build --wheel --no-build-logs --index-strategy unsafe-best-match
   

4. Publish to TestPyPI first:

   uv publish --index testpypi dist/langfuse_mcp-0.x.y-py3-none-any.whl
   

   • Use __token__ as username
   • Use your TestPyPI API token as password

5. Push tag and changes to remote:

   git push origin v0.x.y
   git push
   

6. Publish to PyPI (when ready):

   uv publish dist/langfuse_mcp-0.x.y-py3-none-any.whl
   

   • Use __token__ as username
   • Use your PyPI API token as password