Claude Connects to GitHub and Creates Pull Requests via MCP

Claude works with GitHub through the Model Context Protocol (MCP), Anthropic’s open standard for connecting AI to external systems. Through the GitHub MCP server, the model reads a repository and creates branches, issues and pull requests within the token’s permissions.

What is the Model Context Protocol?

The Model Context Protocol is an open protocol that defines how an AI application gets data and takes actions in external systems. Anthropic published it in November 2024 with SDKs and ready-made servers for GitHub, Google Drive, Slack and databases. Before MCP, every integration was written separately for each model. Now one MCP server works with any compatible client.

How does MCP work?

• Host and client. The app where the model runs: Claude Desktop, Claude Code, an IDE or a company’s own AI assistant.
• MCP server. A program that gives the model access to a system: GitHub, a CRM, a database, files.
• Tools. Actions the model can call: create an issue, find a file, open a pull request.
• Resources and prompts. Context data and ready-made request templates.

A server runs locally or remotely over HTTP. Its permissions are limited to what the token or account it runs under allows.

What can Claude do with GitHub through MCP?

The official GitHub MCP server is maintained by GitHub itself. With it, Claude can:

1. Read code, commit history and open issues in a repository.
2. Create a repository or a branch for a task.
3. Edit files and commit the changes.
4. Open a pull request with a description of the changes.
5. Respond to review comments and update the pull request.

In Claude Code, an agent does the same work in the developer’s terminal, and in Claude on the web, GitHub is added as a connector.

Who else supports MCP?

During 2025 the protocol became an industry standard. OpenAI added support in the Agents SDK and ChatGPT, Google for Gemini, and Microsoft in Windows, Copilot Studio and VS Code. CRM, database, cloud and automation vendors, n8n among them, released MCP servers. In December 2025 Anthropic handed MCP to the Agentic AI Foundation under the Linux Foundation, so the protocol no longer depends on a single company.

How can a business use MCP?

• Development. An agent takes an issue from GitHub Issues, creates a branch and a pull request, and a developer reviews and merges it.
• Sales. An AI assistant sees the client record in the CRM and drafts a reply for the manager.
• Support. A bot answers customers from a knowledge base in Google Drive or Notion.
• Analytics. A manager asks the assistant for database figures in plain language.

We build scenarios like these as AI assistants and chatbots and sales automation.

What are the risks of MCP and how do you reduce them?

• Excess permissions. Scope the GitHub token to the repositories and actions that are needed.
• Prompt injection through data. An issue or file can contain instructions aimed at the model. Have a person review every pull request before merging.
• Unverified servers. Use MCP servers from official sources and keep them updated.
• Action log. Keep a log of tool calls to see what the agent did.

Before connecting AI to business systems, a cybersecurity audit is worth doing.

FAQ

What is MCP in simple terms?

A single way to connect an AI model to a company’s software and data. One MCP server works with Claude, ChatGPT and other compatible clients.

Can Claude create a pull request on GitHub by itself?

Yes, through the GitHub MCP server or Claude Code. Permissions are limited by the token, and merging is best left to a person.

Do I need to code to connect MCP?

For ready-made servers, client settings and an access token are enough. A custom MCP server for a CRM or database is built by a developer.

Is it safe to give AI access to a repository?

Yes, if the token covers specific repositories with minimal permissions and changes go through review. For sensitive systems, start with read-only access.

Who develops MCP now?

Since December 2025 the protocol has been developed in the Agentic AI Foundation under the Linux Foundation, with Anthropic taking part.

Want to connect an AI assistant to your CRM, knowledge base or GitHub? Describe the task, and we will propose the setup, access rights and timing.

Discuss an AI assistantTelegramWhatsApp