GitHub Copilot code review gains MCP and custom skill support
GitHub Copilot code review can now be extended with Model Context Protocol (MCP) servers and custom agent skills, giving every pull request access to external documentation and team-defined review checklists beyond just the diff.
Score breakdown
Teams can encode their own engineering standards and connect external documentation sources once at the repo level, and every subsequent pull request is automatically reviewed against those standards without any per-PR configuration.
- 01GitHub Copilot code review now supports extension via Model Context Protocol (MCP) servers and custom agent skills.
- 02Teams configure MCP servers and enable custom instructions in the repository's code review settings.
- 03The demo uses the Microsoft docs MCP server to surface REST API design guideline violations in PR comments.
GitHub Copilot code review can now be extended using Model Context Protocol (MCP) servers and custom agent skills, moving beyond diff-only analysis to incorporate a team's external documentation and internal standards. The demo walks through a repository containing a Flask back end and an Astro front end, where a new adoption endpoint and adopt button are submitted as a pull request. Before the review runs, the team has configured the Microsoft docs MCP server in the repo settings — the same documentation site where Microsoft publishes REST API design guidelines — and enabled MCP tools alongside custom instructions in the code review settings.
The team's review checklist is stored as a regular markdown file under `.github/skills` in the repository.
The team's review checklist is stored as a regular markdown file under `.github/skills` in the repository. It defines two rules: new endpoints require a test, and error responses must follow Microsoft's documented error envelope format. Because the file is version-controlled with the codebase, every PR automatically picks up the current checklist, and changes to the checklist itself go through the same review process as the rest of the code.
When Copilot is assigned as a reviewer, the session log shows it loading the Microsoft docs MCP server and the checklist skill before running its analysis. The resulting review surfaces multiple findings: one flags a missing test on the new endpoint (mapped to the team checklist rule), another flags an error response that doesn't conform to Microsoft's documented error envelope — language drawn directly from the MCP-connected documentation rather than the diff. Findings from all three sources (external docs, team checklist, and Copilot's built-in analysis) appear together in a single review pass.
Key facts
- 01GitHub Copilot code review now supports extension via Model Context Protocol (MCP) servers and custom agent skills.
- 02Teams configure MCP servers and enable custom instructions in the repository's code review settings.
- 03The demo uses the Microsoft docs MCP server to surface REST API design guideline violations in PR comments.
- 04A team review checklist is stored as a markdown file under `.github/skills` inside the repository.
- 05The checklist defines two rules: new endpoints need a test, and error responses must follow Microsoft's documented error envelope.
- 06Because the checklist is version-controlled, every PR automatically uses the current version of the rules.
- 07A single review pass can surface findings from external MCP documentation, the team checklist, and Copilot's built-in analysis simultaneously.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 8, 2026 · 15:36 UTC. How this works →