M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
G

GitMCP

Community
Free, open-source remote MCP server that turns any GitHub project into an AI-queryable documentation hub
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #26 of 438
Stars
★ 8.5k Very popular
Transport
SSE
Runtime
Hosted (no local runtime) · Node.js
Credentials
No credential needed
License
Apache-2.0
Last commit
Tools
4
67FMRS · C

GitMCP is a free, open-source remote MCP server focused on eliminating AI code hallucinations by giving popular AI coding tools smart, read-only access to any GitHub project's documentation and code. It's simple to set up with no authentication, but is limited to public, read-only content and does not support write operations or private repository access.

Strongest · Documentation 16/20 Weakest · Reliability 9/20

Reliability
9/20
Security and permissions
13/20
Maintenance
15/20
Documentation
16/20
Setup experience
14/20
Why each score
Reliability 9/20
The README describes expected behavior for four tools (fetch_documentation, search_documentation, fetch_url_content, search_code), but no cached manifest, test suite, or CI evidence is available to confirm that declared tools match real behavior. As a remotely hosted service, the init/tool-list handshake cannot be verified from static material alone, and error handling/dependency control are not documented. Per the static calibration rule, scores above 12 require verifiable execution evidence (tests + CI), which is absent here — 9 reflects a plausible happy path with thin edge-case/error-handling evidence.
Security and permissions 13/20
The service requires no authentication, claims no PII collection or query persistence, and respects robots.txt when scraping GitHub Pages — positive signals, and no hardcoded secrets or unsafe destructive defaults were found, so it does not hit a red line. However, all traffic is routed through a third-party hosted intermediary (gitmcp.io), a non-trivial data-boundary concern, and fetch_url_content retrieves arbitrary external URLs referenced in documentation and feeds them to the LLM without any stated limits or confirmation step — a plausible content-injection/SSRF-adjacent risk surface. Overall the main risks are visible but scoping/confirmation is incomplete, keeping the score in the mid-range rather than higher.
Maintenance 15/20
The project has a large star count (8322), is not archived, carries an Apache-2.0 license, and provides a CONTRIBUTING guide plus local dev/test instructions, indicating ongoing community activity. However, the README gives no concrete evidence of release cadence, recent commit history, or a dedicated security-response channel, and the pace of handling 77 open issues cannot be assessed from static material, so it falls short of full marks.
Documentation 16/20
The README is thorough, covering concrete configuration examples for multiple clients (Cursor, Claude Desktop, Windsurf, VSCode, Cline, etc.), an explanation of how the system works, descriptions of all four tools, an FAQ, a privacy section, and local development/debugging steps (MCP Inspector) — layered and reasonably example-rich. Points are deducted because tool parameter schemas/limits (rate limits, timeouts, response size caps) are not disclosed, and troubleshooting guidance is thin, short of the top tier.
Setup experience 14/20
As a zero-install remote service, users only need to add a single URL to their client config, and the README supplies concrete JSON examples for mainstream clients plus self-hosting steps (clone, pnpm install, npm run dev), making the process clear and low-friction. Per the static review rule, however, no executable CI/test evidence exists to verify the configuration flow actually works end-to-end, which caps the score below 15.

Static review · not runListed 2026-08-26

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • Developers using MCP-capable AI coding tools like Cursor, Claude Desktop, Windsurf, or VS Code
  • Users who frequently need up-to-date docs/code from one or many public GitHub projects
  • Teams wanting a zero-install, zero-config remote documentation service

Not for

  • Accessing private repositories or content requiring authentication (GitMCP only serves publicly available content and has no auth)
  • Users who need write access — creating commits, PRs, or issues — since GitMCP only provides read-only documentation/code retrieval
  • AI tools that don't support the Model Context Protocol

Required permissions

  • No GitHub authentication required; only accesses content that is already publicly available
  • Respects robots.txt directives when accessing GitHub Pages sites
  • Runs as a remote cloud service, so the client must be allowed to make outbound network requests (tool calls typically require user approval)

Risks and side effects

  • As a third-party remote service, GitMCP explicitly states it is not affiliated with GitHub or any AI tool vendor, so trust in the hosted service should be evaluated independently
  • The fetch_url_content tool may retrieve third-party external links referenced in a repo's docs, so content quality/safety depends on the original source
  • The maintainers state no personal data or queries are stored, but users relying on external hosted infrastructure should weigh this against their own compliance needs

Setup

Before you start

Runtime：Hosted (no local runtime) · Node.js

  1. Pick the GitMCP endpoint to use: a specific-repository endpoint at https://gitmcp.io/{owner}/{repo} (or {owner}.gitmcp.io/{repo} for GitHub Pages sites), or the dynamic https://gitmcp.io/docs endpoint to let the AI choose the repository at query time.
  2. Add that URL to your AI client's MCP configuration — Cursor, Windsurf, VS Code, and Cline accept the URL directly (url/serverUrl field); Claude Desktop, Augment Code, and Msty AI connect through npx mcp-remote <url> as the launch command.
  3. Save the configuration and restart/reload the client to confirm the MCP server connects.
  4. Ask the AI assistant documentation- or code-related questions about the target repository to trigger tool calls (usually requiring your approval).
  5. To self-host, clone the repository, run pnpm install, then pnpm dev, and test locally with the MCP Inspector over SSE transport.
.cursor/mcp.json 
{
  "mcpServers": {
    "gitmcp": {
      "url": "https://gitmcp.io/{owner}/{repo}"
    }
  }
}

Shown for Cursor. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "gitmcp": {
      "url": "https://gitmcp.io/{owner}/{repo}"
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add --transport http gitmcp 'https://gitmcp.io/{owner}/{repo}'

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

The client's MCP tool list should show tools such as fetch_documentation and search_documentation; ask a question about the repository's docs and confirm the answer is grounded in that documentation to prove the connection works.

Troubleshooting

  1. Verify the URL format matches the intended mode (specific-repository endpoint vs. dynamic gitmcp.io/docs endpoint)
  2. For clients requiring explicit transport configuration, confirm SSE is set correctly (e.g., "type": "sse" in VS Code's mcp.json)
  3. For clients that bridge via `npx mcp-remote` (e.g., Claude Desktop), ensure Node.js/npx is installed and the machine can reach gitmcp.io
  4. Use the MCP Inspector locally with SSE transport to confirm tools return documentation as expected during development

Things to try

Once connected, you can ask your AI assistant things like:

  • Summarize what this project is about and its main features
  • Search the documentation for how to configure the memory feature
  • Fetch the content of this link referenced in the documentation
  • Search the repository code for an example implementation of this feature

Tools 4

fetch_documentation read-only
Retrieves a GitHub repository's primary documentation (llms.txt first, then an AI-optimized version, then README), giving the AI a project overview.
search_documentation read-only
Runs an intelligent search over a repository's documentation for a given query, returning only relevant parts instead of loading everything.
fetch_url_content read-only
Fetches content from external links referenced in a repository's documentation and converts it into an AI-readable format.
search_code read-only
Searches a repository's actual source code via GitHub code search to find implementation examples or technical details.

Use cases

Give an AI coding assistant up-to-date documentation for a library before it generates code, reducing incorrect API usage
Get accurate usage details for niche, new, or fast-changing libraries not covered in the model's training data
Chat directly with a specific repository's documentation from within an IDE via GitMCP's embedded chat
Add a GitMCP badge to your own open-source project's README to show documentation access counts and make it easy for others to connect

Supported clients

Cursor
Claude Desktop
Windsurf
VS Code
Cline
Highlight AI
Augment Code
Msty AI

Listed from the project's documentation, not tested by this site.

Overview

GitMCP is a free, open-source, remote Model Context Protocol server that turns any GitHub repository or GitHub Pages site into a documentation hub AI assistants (Cursor, Claude Desktop, Windsurf, and others) can query. It prioritizes a project's llms.txt file, falls back to an AI-optimized documentation version, and finally to README.md, giving AI tools access to up-to-date, accurate documentation and code so they hallucinate less. GitMCP offers per-repository endpoints (gitmcp.io/{owner}/{repo}) for targeted, secure access, plus a generic dynamic endpoint (gitmcp.io/docs) that lets the AI pick the repository at request time. It requires zero setup — no downloads, installs, signups, or authentication — and can also be self-hosted.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision c487a29895dc Data synced 2026-10-19 Read the FMRS scoring method