GitHub Copilot Code Review v2026: Agent Skills, MCP, Medium Tier
Copilot code review now adapts to your team's tools with agent skills and MCP support. New Medium analysis tier routes complex PRs to deeper reasoning. Configure review depth per repository and control AI compute spend.
TL;DR
- Agent skills and MCP server support bring your team's context into every code review
- New Medium analysis tier routes complex PRs to a higher-reasoning model for deeper inspection
- Admins can now configure review depth per repository and control compute spend
New
- Agent skills for code review — invoke your team's internal tools and standards during reviews by adding SKILL.md files to .github/skills/code-review directory
- MCP server connections — pull context from issue tracking, documentation, service catalogs, and incident tooling directly into reviews via JSON configuration
- Medium analysis tier — routes complex pull requests to a higher-reasoning model for deeper logic analysis, security-sensitive code, and cross-service changes; Low tier remains fast and cost-efficient for straightforward work
- Configurable Actions workflows — control the compute and environment Copilot uses for review
- Shared configuration — platform teams configure once and get consistent behavior across both review and cloud agent
How to Set Up
- MCP servers: Add JSON configuration under repository settings → Copilot → MCP servers, then store authentication tokens under Secrets and variables → Agents
- Agent skills: Create .github/skills/code-review directory with SKILL.md containing context and instructions; existing skills automatically available if relevant
- Review tier: Navigate to repository settings → Copilot → Code review → Review effort level and select Low or Medium from dropdown
Update: Available now in public preview for Copilot Pro, Pro+, Business, and Enterprise users. See GitHub docs for MCP configuration examples and agent skills documentation.
Source: GitHub Changelog
