GitHub MCP Server — 让 AI Agent 直连你的代码仓库
安装 GitHub MCP 服务器，让 Claude Code、Cursor 和 Windsurf 安全地读写你的 GitHub 仓库、Issues 和 PR。粘贴一段 JSON，AI Agent 就能浏览代码、提 PR、处理 Issue。
安装 GitHub MCP Server
{
"mcpServers": {
"github": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-github"],
"env": {
"GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here"
}
}
}
}GitHub MCP Server — Official GitHub AI Integration
GitHub's official MCP server that lets AI assistants manage repos, issues, PRs, Actions, and code search through the Model Context Protocol.
GitHub MCP Server — Manage Repos & Issues from AI Tools
GitHub official MCP server. Connect Claude Code or Cursor to GitHub for managing repos, issues, PRs, and code search via natural language.
GitHub MCP — Repository Management for AI Agents
Official GitHub MCP server that gives AI agents access to repos, issues, PRs, code search, and Actions. Create branches, review PRs, and manage projects via natural language. 5,000+ stars.
Best of MCP Servers — 450 Ranked MCP Servers
A weekly-updated ranked directory of 450 MCP servers across 34 categories with quality scores based on GitHub stars, contributors, and activity. Find the best MCP server for any use case.
LibreChat — Self-Hosted Multi-AI Chat Platform
LibreChat is a self-hosted AI chat platform unifying Claude, OpenAI, Google, AWS in one interface. 35.1K+ GitHub stars. Agents, MCP, code interpreter, multi-user auth. MIT.
Mastra — TypeScript AI Agent Framework
Production TypeScript framework for building AI agents with tool use, workflows, RAG, and memory. First-class MCP support. Deploy anywhere Node.js runs. 9,000+ GitHub stars.
Awesome Claude Code Toolkit — 135 Agents, 176 Plugins
The most comprehensive Claude Code toolkit: 135 agents, 176+ plugins, 35 skills, 42 commands, 20 hooks, 15 rules, and 13 MCP configs. Apache 2.0 license, 1.1K+ GitHub stars.
mcp-agent — Build AI Agents with MCP Patterns
mcp-agent is a Python framework for building AI agents using the Model Context Protocol. 8.2K+ GitHub stars. Implements composable workflow patterns (orchestrator, map-reduce, evaluator-optimizer, rou
Langflow — Visual AI Agent Builder with API
Langflow is a visual platform for building and deploying AI agents as APIs or MCP servers. 146K+ GitHub stars. Multi-agent orchestration, playground, observability. MIT.
Jan — Offline AI Desktop App with Full Privacy
Jan is an open-source ChatGPT alternative that runs LLMs locally with full privacy. 41.4K+ GitHub stars. Desktop app for Windows/macOS/Linux, OpenAI-compatible API, MCP support. Apache 2.0.
Airweave — Context Retrieval Layer for AI Agents, 50+ Integrations
Open-source context retrieval layer connecting AI agents to 50+ apps including Notion, Slack, GitHub, and Jira. Unified search API with MCP support.
FastMCP — Build MCP Servers in Python, Fast
The fast, Pythonic way to build MCP servers and clients. Clean decorator API, automatic type validation, built-in testing, and OpenAPI integration. 24K+ GitHub stars.
Composio — 250+ Tool Integrations for AI Agents
Composio connects AI agents to 250+ tools (GitHub, Slack, Jira, DBs) with managed auth. 15K+ stars. Python/JS SDK, MCP support. AGPL-3.0.
GitMCP — Remote MCP Server for Any GitHub Repo
Free remote MCP server that gives AI agents context from any GitHub project. Eliminates code hallucinations by providing real README, docs, and code. Zero setup. 7.8K+ stars.
LocalAI — Run Any AI Model Locally, No GPU
LocalAI is an open-source AI engine running LLMs, vision, voice, and image models locally. 44.6K+ GitHub stars. OpenAI/Anthropic-compatible API, 35+ backends, MCP, agents. MIT licensed.
Context+ — Semantic Code Intelligence MCP Server
MCP server combining RAG, Tree-sitter AST parsing, and spectral clustering to turn large codebases into searchable feature graphs. 99% accuracy on monorepos. 1,700+ GitHub stars.
为什么要把 GitHub 接入 AI 工作流
The GitHub MCP server is the most-installed AI coding integration of 2026, because it unblocks the single most common AI agent limitation: "your AI can write code, but it can't see the rest of the repository." Once installed, Claude Code, Cursor, or any other MCP-compatible client can list files, read blame, open issues, comment on PRs, and execute searches across your entire GitHub organization — no more pasting snippets into the chat window.
Under the hood, the server speaks the Model Context Protocol — an open standard from Anthropic that lets AI tools securely discover and call external APIs. The GitHub MCP server wraps the GitHub REST and GraphQL APIs into a typed tool surface the model can reason about. You control exactly which repos the AI can touch via your personal access token scopes, and every mutation (comment, PR, issue) is auditable from the GitHub UI.
For teams, the biggest win is context-aware code review. Point the AI at a PR URL and it pulls the diff, reads related files, checks test coverage, and writes a review comment — all without leaving your editor. Combine it with a database MCP server and your AI can go from "analyze this schema change" to "open a PR with the migration" in one conversation. Browse all 16+ GitHub-related MCP assets on TokRepo below.
The GitHub MCP server is the single install that turns an AI coding assistant into a real team member — one that can see the whole repo, not just the file you're editing.
常见问题
What is the GitHub MCP server?+
The GitHub MCP server is an open-source implementation of the Model Context Protocol that exposes GitHub's REST and GraphQL APIs as structured tools an AI agent can call. Once installed in Claude Code, Cursor, or Windsurf, the model can list repos, read files, open PRs, triage issues, run searches, and post review comments — scoped to whatever your personal access token allows.
How do I install the GitHub MCP server in Cursor?+
Create a GitHub personal access token at github.com/settings/tokens (fine-grained is recommended). Open ~/.cursor/mcp.json (create it if missing) and paste the JSON block shown above, replacing ghp_your_token_here with your actual token. Restart Cursor. You'll see a small MCP icon in the chat input once the server connects — that confirms the tools are loaded.
Does the GitHub MCP server work with Claude Code?+
Yes. Claude Code has first-class MCP support. Run `claude mcp add github --env GITHUB_PERSONAL_ACCESS_TOKEN=... -- npx -y @modelcontextprotocol/server-github` and restart Claude Code. You can verify the server is connected with `claude mcp list`. All GitHub tools become available to the agent in every new session.
What GitHub permissions does the MCP server need?+
Minimum: `repo` (for private repos) or `public_repo` (for public-only), `read:org` for org-scoped queries. For PR and issue automation add `write:discussion`. Use a fine-grained token scoped to specific repositories for production use — never give the MCP server broader access than a human teammate would get.
Can I use the GitHub MCP server with self-hosted GitHub Enterprise?+
Yes. The official server accepts a `GITHUB_HOST` environment variable — set it to your GHE hostname (e.g. `github.yourcompany.com`) and it'll route all API calls there. Pair with a self-signed cert trust flag if your enterprise uses a private CA. For air-gapped setups, also check out the <a href="/en/ai-tools-for/self-hosted">self-hosted AI tools</a> directory.
Is the GitHub MCP server safe to use on private repositories?+
It's as safe as the PAT you give it. Best practices: use a fine-grained token scoped to only the repos the AI needs; start with read-only scopes and add write scopes only for specific workflows; rotate tokens on a schedule; never commit the token to git. The MCP server itself doesn't phone home — it runs locally and only calls the GitHub API directly from your machine.