Skip to content
NewsProduct launch

GitHub brings MCP connections and agent skills to Copilot code review in general availability for all paid plans

· by Pondero Newsdesk

The short version

GitHub's Copilot code review can now pull live context from issue trackers, docs, and service catalogs via read-only MCP connections, and apply team-defined standards via SKILL.md files, both generally available as of July 29 for all paid tiers.

GitHub brings MCP connections and agent skills to Copilot code review in general availability for all paid plans

GitHub's AI code reviewer can now pull context from external tools during a review. That capability graduated from public preview to general availability on July 29 across all paid Copilot tiers, per the GitHub Changelog.

What changed

Two features reached GA: agent skills and read-only MCP server connections.

Agent skills let teams supply Copilot with internal coding standards and repository-specific context. A SKILL.md file placed inside a subdirectory under .github/skills/ tells the reviewer what to look for before issuing comments, so it can apply project rules that are invisible to a general-purpose model operating on a diff alone.

MCP server connections let the reviewer fetch context from third-party platforms during a review. Issue trackers, documentation systems, and service catalogs are the documented targets. All MCP tool calls are strictly read-only; the reviewer cannot write to connected services. The GitHub and Playwright MCP servers are enabled by default for all users. Any MCP configurations already set up for the Copilot cloud agent carry over automatically.

Teams that configured either feature during the preview period need no changes. New users configure MCP servers under repository settings, then the Copilot section, then MCP servers. Skills go under .github/skills/. Authentication tokens for MCP connections live under Secrets and variables, then Agents, in repository settings.

One new behavior across both features: attribution. Copilot now labels review comments that drew on a skill or MCP source, so reviewers can tell when a comment came from live tool data versus the model's baseline knowledge.

Separately, the VS Code July 2026 bundle (versions 1.127-1.131, released July 30) added a redesigned Agents window with side-by-side diff view and subagent tracking, multi-chat sessions with per-chat model selection and independent history, Copilot Vision in general availability (images and PDFs in chat), and BYOK model support inside the Agents window.

Why it matters

The prior limitation of automated code review was that the tool only saw the diff. It had no knowledge of your open issues, your API surface documentation, or the architectural decisions your team recorded elsewhere. These two features give the reviewer a direct line to the external systems and internal documents where that context actually lives.

For teams already on GitHub Copilot Business or Enterprise plans, this is included in existing subscriptions. No additional purchase is required. A practical starting point: write one SKILL.md encoding your team's most common review patterns, wire up your primary issue tracker via MCP, and observe whether comment quality shifts before rolling it out to the full organization.

The attribution labels matter for adoption. Engineers are more likely to act on a reviewer comment when they can see it came from a standard the team chose, rather than treating it as a generic suggestion from a tool that doesn't know the codebase.

What to watch next

Whether Cursor and other VS Code-extension competitors ship comparable MCP-backed review integrations is the near-term competitive signal to watch. GitHub's current MCP integration is read-only; writing-mode agent skills (where the reviewer could open an issue or update a document inline) are not yet on the published roadmap.

Sources