GitMCP: A Remote MCP Server That Points AI Assistants at Live GitHub Docs
Put an end to code hallucinations! GitMCP is a free, open-source, remote MCP server for any GitHub project
At a glance
- What is it?
- GitMCP turns any GitHub repository or Pages site into a queryable Model Context Protocol endpoint, so Cursor, Claude Desktop, and other agents read current code and docs instead of guessing. The trade-off is that you are trusting a third-party cloud service with your repository's content.
- Who is it for?
- Adopt GitMCP if you use Cursor, Claude Desktop, Windsurf, or similar MCP-capable tools and need up-to-date documentation for niche or rapidly changing GitHub repositories without local setup. Do not use it if you cannot accept sending repository content to a third-party cloud service, or if your workflow requires offline access.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Activity is slowing. The repository last received commits 6 months ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- idosal/git-mcp8,521 ★ · +142
| Repository | 2026-09-10 | 2026-11-15 |
|---|---|---|
| idosal/git-mcp | 8.4k | 8.5k |
142 Stars in the last 66 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
What GitMCP Actually Fixes
Large language models trained on public code have a knowledge cutoff. When you ask one to use a library released after that cutoff, or a repository with sparse training data, it invents API signatures and examples. GitMCP addresses that by giving the AI a live read-only view of the GitHub repository's documentation and code through the Model Context Protocol. The README's own tagline, 'Stop vibe-hallucinating and start vibe-coding,' frames it as a way to keep the speed of AI-assisted development while grounding answers in the actual source. The intended user is a developer working inside an MCP-capable IDE or chat tool, especially one who depends on libraries that change often or have little representation in training corpora.
The Mechanism: Two Server Flavors and a Smart Search
GitMCP runs as a remote MCP server, not a local process. You configure your client with a URL, and the client connects over HTTP. The server exposes tools that let the AI fetch documentation and code from a specific GitHub repository. The README describes two flavors. The first is a specific repository endpoint like gitmcp.io/{owner}/{repo}, which scopes the AI to one project. The second is a generic endpoint at gitmcp.io/docs, which lets the AI switch between repositories dynamically but requires it to identify the right repo per request. The documentation mentions a 'built-in smart search' that finds relevant content without consuming excessive tokens. That is the core data flow: the AI sends a query, the server searches the target repository's docs and code, and returns only the relevant snippets, reducing both hallucination and token usage.
Setup: Real Commands for Cursor, Claude Desktop, and Others
The README gives concrete configuration steps for several clients. For Cursor, you edit ~/.cursor/mcp.json and add an entry with the URL https://gitmcp.io/{owner}/{repo}. For Claude Desktop, you use npx with the mcp-remote package: the config specifies "command": "npx" and "args": ["mcp-remote", "https://gitmcp.io/{owner}/{repo}"]. Windsurf uses a similar JSON file at ~/.codeium/windsurf/mcp_config.json but with a "serverUrl" key. VSCode's .vscode/mcp.json expects a "type" of "sse" and the same URL. Cline's config lives in a global settings path. The pattern is consistent: no install, no signup, just a URL. That is the advertised 'zero setup' benefit. The README also shows a conversion tool on the landing page that turns a normal GitHub URL into the MCP URL.
The Cloud Dependency and Its Hidden Costs
GitMCP's convenience comes from being hosted. You do not run a server locally, but you also do not control where your repository's content is sent. The README claims the service is 'private' and 'doesn't collect personal information or store queries,' and it says you can self-host. Those are strong claims, and the README has a separate Privacy section to read before relying on them. For a public open-source repository, the privacy risk is low because the content is already public. Also, the generic endpoint relies on the AI's ability to pick the right repository from context, which the README itself flags as a potential source of error.
A Real Limitation: Token Efficiency vs. Full Code Access
The smart search is a double-edged sword. It saves tokens by returning only relevant snippets, but that means the AI never sees the full repository tree unless it issues many queries. For a large codebase, the AI might miss a critical file that the search does not rank highly. The README's example of a three.js scene suggests the tool works well for framework usage, where docs and examples are structured. For understanding cross-file architecture or debugging a subtle integration, a search-based approach may be insufficient. If your work involves exploring unfamiliar code deeply, you might be better served by a local MCP server that gives the AI direct filesystem access.
Alternatives: Local MCP Servers and Official Docs
GitMCP is not the only way to give an AI current code context. The most direct alternative is to run a local MCP server that points at a cloned repository, such as the official GitHub MCP server or a filesystem-based server. The difference in approach is fundamental: GitMCP is remote and stateless, fetching content on demand from GitHub's servers, while a local server reads from your disk, which works for private repositories and offline development. Another alternative is to skip MCP entirely and rely on the AI's built-in web browsing, but that lacks the structured tool interface and token-saving search that GitMCP provides. The choice comes down to whether you prioritize zero setup over control and privacy.
Maintenance, License, and What to Verify
The repository is written in TypeScript and licensed under Apache-2.0, which permits commercial use and modification with attribution. The repository has no GitHub releases, and the last push was on 2026-05-08. The README mentions self-hosting as a feature; check how it is done before relying on it. The hosted service is free, but free services can change terms or disappear. Before adopting GitMCP for a production workflow, check the repository's issue tracker and source code for how the server authenticates requests and whether there are rate limits. Also verify that the SSE endpoint type used by VSCode and Highlight works with your preferred client, as some clients require different transport types.
Editorial conclusion
Adopt GitMCP if you use Cursor, Claude Desktop, Windsurf, or similar MCP-capable tools and need up-to-date documentation for niche or rapidly changing GitHub repositories without local setup. Do not use it if you cannot accept sending repository content to a third-party cloud service, or if your workflow requires offline access. Before adopting, verify that the hosted service's privacy guarantees match your needs and test the generic gitmcp.io/docs endpoint, since it relies on the AI correctly identifying the target repository each time. Self-hosting is available for those who need control.
Frequently asked questions
What is MCP Git?
The project is called GitMCP, not MCP Git. It is a free, open-source, remote Model Context Protocol server that transforms any GitHub project, either a repository or a GitHub Pages site, into a documentation hub an AI tool can read, so the assistant gets current docs and code instead of guessing.
Is GitHub MCP free?
GitMCP is open source and completely free to use. It does not collect personal information or store queries, and you can self-host it. There is no signup and nothing to download, since it runs in the cloud and you only add its URL as an MCP server.
how to install git mcp server
There is nothing to install on the client side. Add the URL to your editor's MCP configuration, for example a `url` of `https://gitmcp.io/{owner}/{repo}` in Cursor's `~/.cursor/mcp.json`. For clients that only speak stdio, such as Claude Desktop, use the command `npx mcp-remote https://gitmcp.io/{owner}/{repo}` instead.
what is git mcp server
It is a remote MCP server that serves a repository's documentation and code. You pick one of three URL shapes: `gitmcp.io/{owner}/{repo}` for a repository, `{owner}.gitmcp.io/{repo}` for a GitHub Pages site, and `gitmcp.io/docs` for a generic server that asks you or infers which repository to read for each request.
is there a git mcp server
Yes, GitMCP is one, and it runs at gitmcp.io rather than on your machine. Pinning the owner and repo in the URL keeps your assistant on the correct project and prevents access to unintended repositories, which is the stated reason to prefer that form over the generic `gitmcp.io/docs`.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/idosal-git-mcp)Star history chart for your README
[](https://hysenlabs.com/projects/idosal-git-mcp)Was this useful?