GitHub MCP Server Review: The Complete Guide to AI-Powered Repository Management
AI Infrastructure Lead

Key Takeaways
- GitHub MCP Server is the official, first-party MCP server from GitHub with 51 tools covering repos, issues, PRs, Actions, and security
- The remote server at
api.githubcopilot.com/mcp/requires zero Docker, zero config -- just OAuth sign-in - Works with 9+ IDEs including VS Code, Cursor, Windsurf, Claude Desktop, and JetBrains
- Completely free and open source (MIT license) -- no paid tiers, no premium features
- 28,300+ GitHub stars make it the most popular MCP server in the ecosystem
- Security concerns exist around prompt injection via issues -- mitigated by Lockdown mode and content sanitization
- Rate limits (5,000 req/hour) can be hit fast during heavy agent usage -- plan accordingly
Table of Contents
What Is the GitHub MCP Server?
The GitHub MCP Server is the official Model Context Protocol server built and maintained by GitHub. It gives AI coding assistants -- Claude, Cursor, Copilot, Windsurf, and others -- direct, structured access to the entire GitHub platform. Repos, issues, pull requests, Actions workflows, code scanning, notifications. All through natural language.
We have been running it daily since October 2025. It has fundamentally changed how we interact with repositories. Instead of switching between the browser, CLI, and editor, we just tell our AI assistant what we need and it handles the GitHub API calls behind the scenes.
GitHub originally released this as a TypeScript-based npm package (@modelcontextprotocol/server-github) in early 2025. That version is now deprecated. The current server is a complete rewrite in Go, built in collaboration with Anthropic, and first shipped as a public preview on April 4, 2025. As of March 2026, it is at version v0.32.0 with 28,300+ stars on GitHub.
The key thing that separates this from community MCP servers: it is a first-party GitHub product. That means direct API access, OAuth integration, enterprise support, and a team of GitHub engineers maintaining it full time. This is not some weekend project that might go stale in six months.
Key Features
With 51 tools across 10 toolsets, the GitHub MCP Server covers almost everything you would do on GitHub. Here are the capabilities we actually use every day.
Repository Management
Create repos, read files, push commits, manage branches, search code across your entire org. The push_files tool lets you commit multiple files in a single operation -- no staging required.
Pull Request Workflows
Create PRs, review diffs, add line-by-line comments, submit reviews, and merge -- all from your AI chat. We use this for automated code reviews that catch issues before human reviewers even look.
Issue Management
Create, update, search, and comment on issues. The real power is bulk operations: "Show me all critical bugs from this week and add a triage label to each one." Done in seconds.
CI/CD and Actions
Monitor workflow runs, debug failed builds, re-run jobs. Ask "Why did the deploy fail?" and get a plain-English summary of the error logs with a suggested fix. This alone saves us 20 minutes per failed build.
Code Security Scanning
List code scanning alerts, review Dependabot findings, check for leaked secrets. The security toolset turns vulnerability management from a manual chore into a conversational workflow.
Copilot Integration
Assign Copilot to work on issues and request Copilot code reviews for PRs. Since v0.32.0, these tools are enabled by default. It is GitHub's AI working with your AI -- surprisingly effective for initial triage.
Dynamic Toolset Discovery
MCP hosts can discover and enable toolsets at runtime based on what the user is asking about. This keeps the tool count manageable and prevents context window bloat -- a real problem with 51 tools.
Read-Only and Lockdown Modes
Read-only mode disables all write operations -- perfect for auditing and exploration. Lockdown mode filters out content from untrusted contributors in public repos. Essential for enterprise security.
How to Set Up the GitHub MCP Server
There are three ways to run the server. We recommend the remote server for most developers -- it takes less than a minute and requires zero local setup.
Remote Server (Recommended -- No Docker Required)
The fastest path. GitHub hosts the MCP server at https://api.githubcopilot.com/mcp/ and handles authentication via OAuth. No tokens to create, no Docker to install.
VS Code (v1.101+): Create or edit .vscode/mcp.json in your project root:
{
"servers": {
"github": {
"type": "http",
"url": "https://api.githubcopilot.com/mcp/"
}
}
}
That is it. VS Code prompts you to sign in with GitHub OAuth automatically. No PAT needed.
Cursor (v0.48+): Edit ~/.cursor/mcp.json:
{
"mcpServers": {
"github": {
"url": "https://api.githubcopilot.com/mcp/",
"headers": {
"Authorization": "Bearer YOUR_GITHUB_PAT"
}
}
}
}
Cursor requires a Personal Access Token in the header. Create one at github.com/settings/tokens with repo scope.
Docker (Local Server)
Use this if you need full toolset control, work behind a corporate firewall, or want to run the server offline.
Claude Desktop: Edit your claude_desktop_config.json:
{
"mcpServers": {
"github": {
"command": "docker",
"args": [
"run", "-i", "--rm",
"-e", "GITHUB_PERSONAL_ACCESS_TOKEN",
"ghcr.io/github/github-mcp-server"
],
"env": {
"GITHUB_PERSONAL_ACCESS_TOKEN": "YOUR_GITHUB_PAT"
}
}
}
}
Claude Code CLI: One-liner to add it globally:
claude mcp add github --scope user \
-e GITHUB_PERSONAL_ACCESS_TOKEN=YOUR_PAT \
-- docker run -i --rm \
-e GITHUB_PERSONAL_ACCESS_TOKEN \
ghcr.io/github/github-mcp-server
Configure Toolsets (Optional)
By default, the server enables repos, issues, pull_requests, and copilot toolsets. If you want Actions or security scanning, enable them explicitly.
# Enable specific toolsets via environment variable
GITHUB_TOOLSETS=repos,issues,pull_requests,actions,code_security
# Or enable everything
GITHUB_TOOLSETS=all
# Read-only mode (disables all write operations)
GITHUB_READ_ONLY=1
We recommend starting with the defaults and only adding toolsets when you need them. Loading all 51 tools can overwhelm some AI models -- Cursor, for example, only passes the first 40 tools to its agent.
Pricing and Rate Limits
This is the part we like best. The GitHub MCP Server is completely free. No freemium tricks, no premium tiers, no enterprise upsells on the server itself. The only costs are whatever you already pay for GitHub and your AI assistant.
Free (All Users)
$0
forever
- MIT open source license
- 51 tools across 10 toolsets
- 5,000 API requests/hour
- Remote server with OAuth
- Docker and source builds
- All IDE integrations
GitHub Pro
$4
/month
- Everything in Free
- 5,000 API requests/hour
- Advanced code security
- Protected branches
- Draft PRs
- Same MCP access
Enterprise Cloud
$21
/user/month
- Everything in Pro
- 15,000 API requests/hour/repo
- Custom MCP endpoint
- Lockdown mode
- SAML/SSO
- Audit log streaming
The rate limit is the real constraint. At 5,000 requests per hour, you might think that is plenty. It is not. During a heavy AI agent session -- say, reviewing a large PR, searching code, and managing issues -- we have seen 2,000+ requests consumed in under 15 minutes. GitHub issue #933 documents users hitting the cap regularly.
Our advice: use conditional requests, avoid polling, and scope your toolsets to only what you need. Enterprise Cloud's 15,000/hour limit is the real answer for teams using this at scale.
Pros and Cons
Pros
- Official first-party product -- maintained by GitHub engineers, not a community side project
- Zero-config remote server -- OAuth sign-in, no Docker, no tokens for VS Code users
- 51 tools covering the full API surface -- repos, PRs, issues, Actions, security, notifications
- Completely free -- MIT license, no paid tiers, works on free GitHub accounts
- Enterprise-grade security -- Lockdown mode, content sanitization, scope filtering
- Works with every major IDE -- VS Code, Cursor, Windsurf, Claude Desktop, JetBrains, and more
- Active development -- monthly releases, responsive to community issues
Cons
- Rate limits hit fast -- 5,000 req/hour sounds generous until your AI agent burns through 2,000 in 15 minutes
- Prompt injection risk -- malicious GitHub issues can hijack AI agents (mitigated but not eliminated)
- Tool overload -- 51 tools can confuse AI models, Cursor drops tools beyond the first 40
- No OAuth in Windsurf -- PAT-only authentication limits the zero-config experience
- Docker image pull issues -- occasional ghcr.io auth failures require manual token cleanup
- Remote server has limited toolset config -- less flexibility than the local Docker server
- GitHub-only -- no GitLab, Bitbucket, or Azure DevOps support
Alternatives Comparison
We tested the top alternatives head to head. Here is how they stack up.
Our take: If you only use GitHub, the official server is the clear winner. It has the deepest integration, best security features, and most active development. GitKraken MCP is the better choice if your team uses multiple platforms like GitLab or Bitbucket alongside GitHub. GitMCP is fine for read-only documentation use cases but cannot replace either for real workflow automation.
Frequently Asked Questions
What is the GitHub MCP Server?
The GitHub MCP Server is an official, first-party server built by GitHub that implements the Model Context Protocol standard. It gives AI coding assistants like Claude, Cursor, and Copilot direct access to GitHub repositories, issues, pull requests, Actions workflows, and code security features through natural language conversation.
Is the GitHub MCP Server free?
Yes, completely. The server is open source under the MIT license. The GitHub API is also free within rate limits (5,000 requests/hour with a PAT). The remote server at api.githubcopilot.com is free for all GitHub users regardless of plan. There are no paid tiers or premium features.
How do I set up the GitHub MCP Server in VS Code?
Create a .vscode/mcp.json file with the server type set to "http" and the URL set to "https://api.githubcopilot.com/mcp/". VS Code handles OAuth authentication automatically. No Docker, no PAT, no manual configuration required.
Is the GitHub MCP Server safe to use?
Mostly yes, with caveats. A prompt injection vulnerability was found in May 2025 where malicious issue content could hijack AI agents. GitHub has since added Lockdown mode, content sanitization, and scope filtering. Best practice: use fine-grained PATs scoped to specific repos and enable Lockdown mode for public repo work.
How many tools does the server have?
The remote server exposes 51 tools across 10 toolsets: repos, issues, pull_requests, actions, code_security, copilot, secret_protection, notifications, stargazers, and experiments. You can enable or disable toolsets individually to keep the tool count manageable for your AI model.
What is the difference between the remote and local server?
The remote server at api.githubcopilot.com requires no local setup and supports OAuth. The Docker-based local server gives more toolset control via environment variables and works behind firewalls. Both provide the same core 51 tools.
Can I use it with Claude Desktop?
Yes. Claude Desktop uses the Docker-based local server. Add the configuration to your claude_desktop_config.json file with your GitHub PAT and the Docker image ghcr.io/github/github-mcp-server. Pull the image first with docker pull ghcr.io/github/github-mcp-server.
What are the rate limits?
With a PAT, 5,000 requests per hour. Enterprise Cloud gets 15,000/hour per repo. Heavy AI agent usage can burn through these fast -- we have seen 2,000+ requests in 15 minutes during code review sessions. Use conditional requests and limit toolsets to conserve your budget.
Final Verdict
The GitHub MCP Server is the best MCP server we have tested in 2026. That is not hyperbole -- it is the most complete, most actively maintained, and most widely supported MCP integration available today.
We have been running it with Claude Code and Cursor across five active repositories for the past five months. It has eliminated our need to context-switch between the browser and editor for 90% of GitHub operations. PR reviews, issue triage, CI debugging, code search -- all handled through natural language without leaving the IDE.
The remote server with OAuth is genuinely impressive. Three lines of JSON and you have full GitHub access in VS Code. No Docker, no tokens, no configuration files to manage. GitHub made the right call shipping this as a hosted service alongside the self-hosted option.
The security story is solid but not perfect. The prompt injection vulnerability from May 2025 was a serious architectural issue, and while Lockdown mode and content sanitization have addressed the most dangerous attack vectors, the fundamental problem -- broad PATs giving AI agents access to repos they should not touch -- is still something you need to manage yourself with fine-grained tokens.
Rate limits are the other real pain point. 5,000 requests per hour sounds like a lot until your AI agent decides to fetch every file in a monorepo. You will learn to be specific with your prompts, and that is actually a good habit to build.
Bottom line: If you write code and use GitHub, install this today. It is free, it takes 60 seconds to set up, and it will change how you interact with your repositories. We give it a 4.5 out of 5 -- docking half a point for the rate limit constraints and the still-evolving security model.
4.5 / 5
Our Rating
Build Something With MCP?
We review MCP servers, AI coding tools, and developer productivity software every week. If you have built an MCP server or AI tool, we want to hear about it.
Submit Your Tool for ReviewRecommended AI Tools
RepoClip
RepoClip turns your GitHub repo into a cinematic demo video in 5 minutes. Uses Gemini for code analysis and OpenAI for narration. Free tier is limited but the concept is unique. Rating: 4.0/5.
View Review →Relia
Relia is a Chrome extension that catches broken logic in AI-generated code before your users do. Zero setup, real-time analysis, but pricing is opaque and it is browser-only. Rating: 3.8/5.
View Review →Droidrun
We tested Droidrun for mobile automation. It hit 91.4% on AndroidWorld at just $0.075/task — 12x cheaper than vision-based competitors. The accessibility API approach is smart, but iOS support and cloud platform are still developing. Rating: 4.2/5.
View Review →Adobe Firefly
Updated March 2026 · 12 min read · By PopularAiTools.ai
View Review →