Summary
Read a repository you have not cloned.
It gives an assistant the four moves you would make in a browser tab — see the tree, open the file, search the code, compare two branches — without leaving the conversation or waiting on a clone.
What it is
An MCP server that exposes GitHub repository contents and metadata to a client, with optional local cloning and caching so repeated reads of the same repo do not re-fetch everything.
What you get
- A repository summary, with stars, forks, description and last-updated when you ask for metadata
- The whole directory tree, rendered as an ASCII tree rather than a flat path dump
- The contents of specific files you name, on the branch you name
- Content search inside a repository, so a symbol or string is located without cloning it yourself
- A diff between any two branches or commits
- Text or structured JSON output, and progress notifications on long-running reads
Requirements
Npx @trishchuk/github-explorer-mcp for stdio, a separate entry point for HTTP/SSE mode, or the Docker image on port 3000. The published npx and Docker setups take no key or token.
Setup effort
One command — npx -y @trishchuk/github-explorer-mcp
