GitHub's MCP (Model Context Protocol) server enables AI agents to interact with GitHub repositories, issues, pull requests, and files. Sam Marorrow (lead developer) and Toby Padilla (principal PM) demonstrate how to install the server from GitHub's MCP registry and use it with VS Code's agent mode to automate tasks like listing issues, assigning coding agents, creating PR summaries with mermaid charts, and pulling files from repositories.
Sam Marorrow (lead developer of GitHub's MCP server) and Toby Padilla (principal PM for MCP initiatives and member of the MCP steering committee) introduce the Model Context Protocol. MCP enables AI to fetch context the model wasn't trained on (post-training data, private information) and create side effects in external systems (files, repos, designs).
MCP is about 11 months old, created by Anthropic in November 2024. Toby describes it as "more of a social innovation than a technical innovation" because function calling APIs already existed, but MCP standardized them and Anthropic bootstrapped the community with 30 high-quality reference servers.
The original GitHub MCP server was part of Anthropic's reference servers and became the most popular. GitHub took ownership and rewrote it in Go (from TypeScript). During launch week, it became the most popular open source project on all of GitHub, demonstrating pent-up demand to drive GitHub's platform with agents.
Installing from MCP Registry: Sam demonstrates one-click installation from github.com/mcp into VS Code, noting that VS Code is "the most comprehensive MCP host that exists" with better protocol coverage than Anthropic's hosts.
Listing Issues: A simple prompt to "list some issues from a repo" immediately triggers the MCP server's list_issues tool.
Assign Coding Agent Prompt: MCP supports pre-baked prompts (slash commands) that enable complex agentic workflows. The "assign coding agent" prompt asks for a repository, analyzes open issues, decides which ones are suitable for automation, and assigns coding agents while filtering out low-quality issues.
Creating Summary Issues: The demo shows asking the agent to pull down recent PRs and generate an issue celebrating the work, complete with mermaid diagrams.
Resource Access: The MCP server provides access to any file in any repository you have access to, in any revision or branch.
Copilot coding agent has both GitHub MCP and Playwright MCP installed by default. Playwright enables browser automation for taking screenshots, so agents can show before/after visuals of their work without requiring local builds. This creates a quick round-trip for feedback on visual changes.
With over 1,000 forks and high issue volume, Toby uses the MCP server to summarize issues, classify by priority, and cluster by topic - scaling management of high-volume community feedback.
The GitHub MCP server supports read-only mode, selective tool sets, secret scanning, enterprise integration (GHC), and is included by default in Copilot coding agent. GitHub launched github.com/mcp as a registry of high-quality partner servers, with plans to allow self-publication.
"MCP is more of a social innovation than a technical innovation because there was pre-existing function calling APIs, but they weren't getting a lot of traction."
-- Toby Padilla
"VS Code is actually the most comprehensive MCP host that exists. It has more support for the MCP protocol than any of Anthropic's hosts."
-- Sam Marorrow
"The week that we launched it, it was actually the most popular open source project on all of GitHub."
-- Toby Padilla
"From very basic prompts, it's able to generate some really advanced things... every time I see this, it blows my mind."
-- Sam Marorrow
"Once you get addicted to using the MCP and the various agentic tools that you can connect with it, you just walk around... I use GitHub's mobile app to edit websites when I'm walking in the park."
-- Sam Marorrow