The workflow it wants is summary first, then narrow: take the overview, and pull the tree or the full content only once you know which part of the repository the question is about. Include and exclude patterns are how a big repository stays inside a context window.
One tool over the gitingest library. Point it at a repository URL or a local path and it ingests on demand, returning a summary, a file tree, the content, or all three.
- A repository ingested from a URL or a local path, with no separate ingestion step
- A summary, a tree, the content, or all of them, chosen per call
- Include and exclude patterns, so a large repository is narrowed before it is read
- A maximum file size, and a token ceiling that truncates the response
- A specific branch analysed, and the output written to a file path
- The tree and content also reachable as resources after a call
Uvx to run the package. Public repositories need nothing else; a GitHub token is read from the environment for private ones, which depend on a gitingest release the author notes had not reached PyPI at the time of writing.
One command plus a key — uvx trelis-gitingest-mcp, then supply credentials
