repomuse
v1.0.0
Published
MCP server — fetch, parse, and understand any public GitHub repo as inspiration context for your CLI AI agent.
Maintainers
Readme
RepoMuse MCP Server
Read to understand. Not to copy.
RepoMuse is an open-source MCP (Model Context Protocol) server that lets any MCP-compatible AI agent (Claude Code, Aider, Cursor, etc.) fetch, parse, and understand any public GitHub repository as structured inspiration context — without leaving your CLI session.
Why RepoMuse?
When building something new, developers often want to know "how did X solve this problem?" — but they have to:
- Open GitHub in a browser
- Browse the tree manually
- Copy-paste código into the AI chat
- Explain what they're looking at
RepoMuse eliminates all of that. Just say:
"Study how fastapi handles dependency injection and help me do something similar."
...and your agent just works.
How It Works (Safety First)
RepoMuse is completely stateless and secure. It interacts exclusively with the public GitHub REST API (api.github.com and raw.githubusercontent.com).
- No Git required: It does not run
git cloneor pull files onto your local hard drive. - No Auth needed: For public repositories, no GitHub token or login is required.
- Read-only: It cannot modify any code locally or remotely. It simply fetches text over HTTPS to feed into your AI's context window.
Tools
| Tool | What it does |
|---|---|
| fetch_repo_structure | Get a clean, filtered file tree of any public repo. A great first step to orient the agent. |
| fetch_repo_files | Read the actual source code of specific files. If no files are specified, uses Smart Auto-Select to intelligently pull the most critical architectural files (ignoring configs/docs). Token-safe, 80k char ceiling. |
| summarize_repo_patterns | Fetch structure, manifests, and auto-selected source entry points to return an analysis prompt. Works for JS/TS, Python, Go, Rust, Ruby. |
Quick Setup
RepoMuse is hosted universally on NPM and runs as a standalone stdio MCP server.
The universal execution command to load this server on any OS is:
npx -y repomuseHere's how to plug that command into the most popular AI environments on Mac, Windows, or Linux.
Cursor / Windsurf / Codeium
- Open your IDE Settings
- Locate the MCP section
- Click Add New MCP Server
- Configure as follows:
- Name:
repomuse - Type:
command - Command:
npx -y repomuse
- Name:
Aider
Run aider directly while auto-attaching the MCP via flag:
aider --mcp "npx -y repomuse"Claude Desktop
The easiest way to install this for Claude Desktop is to open the configuration file directly from the app:
- Open Claude Desktop.
- Go to Settings (Click "Claude" in the Mac menu bar, or the
≡menu in Windows). - Under the Developer tab, click Edit Config.
Alternatively, you could manually create/edit the claude_desktop_config.json file in:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json(if supported)
Add the following to your config file:
{
"mcpServers": {
"repomuse": {
"command": "npx",
"args": ["-y", "repomuse"]
}
}
}Claude Code (CLI)
claude mcp add repomuse -- npx -y repomuseOptional: GitHub Token
RepoMuse works out of the box for all public repos — no auth required.
GitHub's unauthenticated API limit is 60 requests/hour per IP. If you use RepoMuse frequently, supply a token to raise this to 5,000 req/hour:
{
"mcpServers": {
"repomuse": {
"command": "npx",
"args": ["-y", "repomuse"],
"env": {
"GITHUB_TOKEN": "ghp_your_token_here"
}
}
}
}Create a token at github.com/settings/tokens — no special scopes needed for public repos.
Example Usage
Once connected, you can say things like:
- "Use repomuse to fetch the structure of https://github.com/expressjs/express"
- "Fetch the router and middleware files from https://github.com/fastapi/fastapi"
- "Summarize the architecture of https://github.com/supabase/supabase"
Token Safety
RepoMuse is designed to be credit-conscious:
| Limit | Value | |---|---| | Directory listing cap | 500 files | | Auto file fetch (smart-select) | 10 core source code files | | Explicit file fetch | 15 files | | Total output ceiling | 80,000 chars (~20k tokens) | | Per-file ceiling (summarize) | 20,000 chars |
All limits include clear truncation warnings so the agent knows when it's been cut off.
Requirements
- Node.js 18+
- Public GitHub repository
- Any MCP-compatible agent
License
MIT — use it, fork it, contribute back.
