Table of contents
GitHub isn’t just for developers. If your marketing website runs on Astro, Next.js, or Hugo, your files live on GitHub. If your documentation uses GitBook or Mintlify, it’s backed by GitHub. If your team uses Obsidian with git sync, or manages projects in GitHub Issues — you’re already a GitHub user, whether you realise it or not.
The GitHub MCP Server changes what that means. It gives AI agents like Claude direct access to your repositories, issues, pull requests, and projects — turning GitHub from something your dev team manages into something you can interact with through natural conversation.
And with updates through to August 2026 — including support for the new stateless MCP specification — the server keeps getting more powerful.
What Is the GitHub MCP Server?
The GitHub MCP Server is an official, GitHub-hosted Model Context Protocol server that connects AI tools directly to GitHub’s platform. Instead of navigating GitHub’s interface or asking a developer for help, your AI agent can read code files, manage issues, create pull requests, track projects, and monitor workflows — all through a single connection.
It works with Claude Desktop, Cursor, VS Code, Windsurf, and most other AI coding tools. Authentication is handled through OAuth (one-click browser approval) or a Personal Access Token, and GitHub hosts the server for you — no infrastructure to manage.
The server currently has over 30,000 stars on GitHub, making it one of the most popular MCP servers available.
What’s Changed Since Launch
The January 2026 update introduced several features that make the server more practical for non-technical users, and two further updates since then have kept pace with the wider MCP ecosystem.
Consolidated Projects Tools
GitHub replaced a sprawl of project-related tools with three unified ones:
- projects_list: Retrieves projects for users, organisations, or repositories
- projects_get: Fetches detailed project information with fields and items
- projects_write: Enables creation, updates, and management of project items
This reduced context window usage by roughly 50%, meaning your AI agent can do more within a single conversation without hitting limits. The system now detects owner types automatically, so you don’t need to specify whether something belongs to a user or an organisation.
OAuth Scope Filtering
The server now intelligently filters which tools are available based on your authentication method. If you’re using a classic Personal Access Token, the server detects your token’s scopes and hides tools you don’t have permission to use. This prevents confusing errors when you try something your token doesn’t allow.
Insiders Mode
An opt-in experimental mode gives early access to preview features and unreleased capabilities. You can enable it through a configuration header or by using the /insiders URL, and revert instantly if needed.
HTTP Server Mode
Enterprise teams can now run the server in HTTP mode on configurable ports, with per-request OAuth tokens. This is primarily useful for larger organisations with custom infrastructure requirements.
Copilot Coding Agent Improvements
New tools for monitoring and managing GitHub’s Copilot coding agent, including get_copilot_job_status for tracking progress and support for custom base branches in pull request creation. This is relevant if your team uses Copilot to automate code changes across your marketing site.
Stateless MCP Spec Support (23 July 2026)
Ahead of the MCP protocol’s move to a stateless core on 28 July 2026, GitHub updated its server to support the new specification early. The session handshake and the Mcp-Session-Id header are gone, so requests no longer need to be pinned to the server instance that created a session. In practice this means faster connections and easier horizontal scaling for hosted deployments, with nothing required on your end.
v1.7.0: Server-to-Server Auth and Faster Projects (28 July 2026)
The server also picked up the underlying MCP Go SDK’s v1.7.0 release, adding server-to-server authentication for GitHub Apps over stdio (useful for enterprise and cloud workloads), batched update_project_items writes via GraphQL for faster bulk project updates, and further lockdown mode improvements.
Why Marketers Should Care About GitHub
Most marketing teams interact with GitHub without realising it. Here are the most common ways.
Marketing Websites
Static site generators like Astro, Next.js, Hugo, and Gatsby all store their content as files in a GitHub repository. Every blog post, landing page, and product page is a file that gets committed, reviewed, and deployed through GitHub.
With the GitHub MCP Server, you can ask your AI agent to:
- Find and read the content of any page on your site
- Create a new blog post by adding a file to the right directory
- Update copy on an existing landing page
- Review what changed on the site recently by looking at commit history
- Open a pull request with your changes so a developer can review before it goes live
This means you can make content changes without learning git commands or asking your dev team for help with every small update.
Documentation Sites
Tools like GitBook, Mintlify, ReadMe, and Docusaurus all sync with GitHub repositories. Your product documentation, help centre, and API docs are just files in a repo.
The MCP server lets you:
- Search across your entire documentation for outdated information
- Update docs directly through your AI agent
- Track which documentation pages have changed recently
- Create new documentation pages in the correct format and location
Note-Taking and Knowledge Management
Obsidian, one of the most popular note-taking tools for technical and non-technical users alike, supports git-based sync through community plugins. If your team stores meeting notes, strategy documents, or research in Obsidian with git backing, the GitHub MCP Server gives your AI agent access to that knowledge base.
Project Management
GitHub Issues and GitHub Projects aren’t just for bug tracking. Many teams use them for marketing project management, content calendars, and campaign planning. With the new consolidated projects tools, your AI agent can:
- Create and update project items
- Move tasks between columns or statuses
- List all items in a project with their current state
- Add new issues for content ideas, campaign tasks, or website updates
Other Common Use Cases
Content calendars: Store your editorial calendar as a GitHub Project and let your AI agent manage it, add new ideas, update statuses, and track deadlines.
Design system documentation: If your design system lives in a repo (as most do), the MCP server lets you search components, read usage guidelines, and track changes.
Marketing automation configs: Many marketing tools (email templates, A/B test configs, analytics setups) are stored as code. The MCP server gives you visibility into these without needing to understand the code itself.
Change tracking and auditing: Every change on GitHub is tracked with who made it and when. Your AI agent can review recent changes to your site or docs, giving you a clear picture of what’s been updated.
Release monitoring: If your product team uses GitHub releases, you can track new versions and features to inform your marketing messaging and content planning.
How the GitHub MCP Server Compares to Other Options
GitHub’s server is one of several repository-focused MCP servers. Here’s how it stacks up against a self-hosted generic git MCP server and GitLab’s own MCP server:
| GitHub MCP Server | Self-hosted / generic git MCP | GitLab MCP Server | |
|---|---|---|---|
| Official support | Yes, built and maintained by GitHub | No, community-maintained | Yes, built and maintained by GitLab |
| Hosting | Hosted by GitHub, nothing to run | You run and maintain it (Docker or a local process) | Hosted by GitLab |
| Breadth of toolsets | Repos, issues, PRs, projects, Actions, code security, and more | Varies, usually a narrow subset | Repos, issues, merge requests, CI/CD pipelines |
| Auth | One-click OAuth or Personal Access Token | Typically API token only | OAuth or Personal Access Token |
| Works with | GitHub only | Any git remote, depending on implementation | GitLab only |
The main limitation is that GitHub’s server only works with GitHub. If your team uses GitLab or Bitbucket, you’ll need a different solution.
Getting Started
Setting up the GitHub MCP Server is straightforward:
- Choose your AI tool: Claude Desktop, Cursor, VS Code, or any MCP-compatible client
- Connect via OAuth: The remote server at
https://api.githubcopilot.com/mcp/supports one-click OAuth authentication — just approve access in your browser - Start asking questions: Try simple requests first, like “show me the recent commits on our website repo” or “list all open issues labelled as content”
For teams that want more control, you can use a Personal Access Token instead of OAuth, and configure which toolsets are available to limit what the AI agent can access.
Practical Tips for Marketing Teams
Start with read-only access. Before letting your AI agent create pull requests or modify issues, spend time using it to search and read your repositories. Get comfortable with what’s there.
Use pull requests for content changes. Rather than pushing changes directly, have your AI agent create pull requests. This gives your dev team a chance to review before anything goes live.
Set up toolsets thoughtfully. You probably don’t need access to code security scanning or GitHub Actions. Configure the server to only expose the tools your team actually needs — repos, issues, and pull requests are a good starting point.
Combine with other MCP servers. The GitHub MCP Server works well alongside other servers. Use it with Firecrawl for competitive research, Notion for strategy docs, or your analytics MCP server for performance data — all in the same AI conversation. Browse the full landscape on our MCP servers hub.
Should Your Marketing Team Use It?
The GitHub MCP Server makes most sense for marketing teams that:
- Manage a website built with a static site generator (Astro, Next.js, Hugo, Gatsby)
- Maintain product documentation on a git-backed platform
- Use GitHub Issues or Projects for any part of their workflow
- Want to make small content changes without depending on developers
- Need visibility into what’s changing on their website or docs
If your marketing stack doesn’t touch GitHub at all, this isn’t for you. But if any of your content, documentation, or project management flows through GitHub — even indirectly — the MCP server removes a layer of friction that’s been slowing you down.
The 2026 updates — from consolidated projects tools through to stateless MCP support and server-to-server auth — make this more accessible and more production-ready than ever before.
About Growth Method
If your marketing site, docs, or content calendar already live on GitHub, you’re closer to running an agentic marketing team than you might think. Growth Method is the agentic marketing platform for B2B teams: plan, launch, and analyse campaigns with pre-built AI agents that work from your live marketing data, for people and agents alike. GitHub is one of the integrations we connect to out of the box, so your AI agents can read commit history, ship content changes, and track project status in the same conversation where they plan and measure the campaign itself.
Get started: first month free, no credit card required.
Frequently asked questions
Is the GitHub MCP Server free? Yes. The server itself is free and open source (MIT-licensed). You only pay for the GitHub plan and AI tooling you already use; there’s no separate charge for the MCP server.
Does it work with GitLab or Bitbucket? No. The GitHub MCP Server only works with GitHub repositories. If your team uses GitLab or Bitbucket, you’ll need their equivalent MCP server instead.
What changed in the July 2026 MCP spec update? GitHub added support for the MCP protocol’s stateless core ahead of its official 28 July 2026 release, removing the session handshake so hosted deployments can scale more easily across multiple instances.
Do I need a Personal Access Token or can I use OAuth? Either works. OAuth gives one-click browser approval and is easiest for most users; a Personal Access Token gives more granular control over which toolsets are exposed.
Can I limit what the AI agent can access? Yes. You can configure which toolsets are available (for example, restricting access to repos, issues, and pull requests) and the server automatically hides tools your token’s scopes don’t permit.