Summary
Two tools, and together they answer "how does this project work" in one exchange.
The pattern is always the same: get the tree, pick the handful of files that matter, read them. Because clones are keyed by a hash of the URL, follow-up questions about the same repository reuse the checkout instead of pulling it down again — which is what makes the back-and-forth bearable.
What it is
A two-tool server that clones a GitHub repository into a temporary directory and reads it — the structure, then whichever files you name.
What you get
- github_directory_structure — a tree view of a repository from its URL, with .git skipped
- github_read_important_files — the contents of the file paths you list, returned as a path-to-content mapping
- Repositories cached by a hash of their URL, so asking a second question about the same repo does not clone it again
- Temporary directories cleaned up after each call
Requirements
Uvx, run straight from the repository with --from and the git URL. Python 3.8 or higher; it uses fastmcp and gitpython underneath. Nothing to authenticate for public repositories.