GitHub Copilot code review ships with Model Context Protocol generally available

GitHub turns on MCP inside Copilot code review for every paid tier

GitHub just made Model Context Protocol server connections and agent skills generally available inside Copilot's automated code review, for all Pro, Pro+, Business, and Enterprise users at once 1. Two MCP servers, GitHub and Playwright, are switched on by default. Any MCP configuration already set up for Copilot's cloud agent carries over to code review automatically. Teams who configured these features during public preview need to change nothing 1.

Two capabilities ship in this release. MCP server connections pull context from third-party platforms, including issue trackers, documentation systems, and service catalogs, directly into the review 1. Agent skills extend the reviewer with repository-specific standards: drop a SKILL.md file under .github/skills, and the reviewer applies those instructions during its analysis 1. The release also adds attribution. Copilot code review now labels which comments were produced using agent skills or MCP context, so each finding traces back to the tool that generated it 1.

The governance decision that makes this safe to ship by default is easy to miss. Every MCP tool call performed by Copilot code review is limited to read-only operations 1. Your reviewer can read a ticket to understand what a change is supposed to accomplish, but it cannot update that ticket, close it, or modify any connected system. An AI reviewer with write access to external platforms would require sign-off and careful opt-in configuration before a team would knowingly enable it. A read-only reviewer is a consumer of context, not an actor inside your infrastructure. That distinction is what lets GitHub switch MCP servers on across every paid tier simultaneously rather than gating each connection behind a security review.

What changes today is distribution, not capability. MCP, the open-source standard for connecting AI applications to external systems, already had support from Claude, ChatGPT, Visual Studio Code, and Cursor 2. Turning servers on by default inside Copilot code review shifts MCP adoption from a deliberate architectural choice teams make to the assumed baseline they start with. The protocol is no longer something builders opt into. It is the starting configuration, and teams would need to actively disable servers to avoid it. That reframes what adoption means for a standard now embedded in every Copilot code review.

For teams, the practical step is immediate. Check which MCP servers are active in your repositories: navigate to Settings, then Copilot, then MCP servers 1. GitHub and Playwright servers are already active, which means your code reviewer has MCP connections you may not have configured yourself. If you set up MCP for Copilot's cloud agent, those configurations now extend into code review, and any third-party platform you connected there is visible to your reviewer. Store new authentication tokens under Settings, Secrets and variables, then Agents 1. Audit what your code review can see before your next pull request.

The read-only constraint is the feature that lets GitHub ship this broadly. It is also the line to watch. If a future release loosens it to allow write operations, even optionally, the governance calculus changes entirely, and the convenience of default-on becomes a very different trade-off.