Player not loading? Watch on YouTube
Matt Soucoup demonstrates GitHub MCP Server with GitHub Copilot in VS Code. He explains how MCP exposes API functions as tools with natural language descriptions that an AI agent can use to choose and call them.
The setup distinguishes a server running locally in Docker from GitHub's hosted HTTP server. Soucoup chooses the hosted option and installs it for the current workspace. The resulting configuration points to api.githubcopilot.com/mcp. This walkthrough uses a remote service; it does not demonstrate running a model on the user's hardware.
In a conference session voting app, Soucoup asks Copilot to list repository issues, find those assigned to him and rank open work by high impact and low effort. He then asks the coding assistant to fix a leaderboard sorting bug and open a pull request. The demonstrated result orders sessions by vote count.
The workflow continues with a Copilot code review in the cloud. Back in VS Code, Copilot retrieves the feedback through MCP, addresses one actionable comment and reports that it pushed the fix and resolved the thread. Soucoup also requests a performance analysis that produces three new issues, plus a search for recurring patterns across unassigned issues. He presents these examples as ways to reduce routine GitHub work; the results belong to this demonstration rather than a general guarantee of accuracy.
