Copilot code review can work with more project-specific context

GitHub Copilot code review can now use Agent Skills and Model Context Protocol connections as generally available features. Released on July 29, 2026, the update gives development teams a way to bring repository-specific instructions, internal standards, and external context into automated reviews, making Copilot more adaptable to the way individual projects are actually built and maintained.

{getToc} $title={Table of Contents}

Copilot code review can work with more project-specific context

Code review becomes more useful when it understands more than the code visible in a pull request. Agent Skills allow teams to provide Copilot with instructions and context tailored to a repository or organization, while MCP connections can bring information from external systems into the review process.

For frontend teams, that creates room for reviews that better reflect internal conventions rather than relying only on general coding knowledge. A project could use its own documented standards, workflow instructions, or specialized tooling as additional context when Copilot analyzes a change.

{getCard} $type={post} $title={More useful reading}

Agent Skills add repository instructions to the review process

Agent Skills can define specialized instructions, scripts, and resources that Copilot loads when they are relevant to a task. For code review, teams can create a skill-specific directory under .github/skills and include a SKILL.md file containing the context and instructions Copilot should follow.

This can be useful for repositories with established development conventions. A frontend project, for example, may have its own component patterns, accessibility expectations, naming rules, or implementation practices. Skills provide a structured way to make those expectations available during Copilot-assisted review instead of depending entirely on generic recommendations.

MCP connections bring external context into code review

MCP server connections extend the review beyond information stored directly in the repository. Copilot can use configured MCP tools to access context from platforms such as issue trackers, documentation systems, service catalogs, and other sources already used by a development team.

Repository administrators can configure MCP servers from the repository's Copilot settings. Existing MCP configurations previously used with the Copilot cloud agent continue to work with code review, and GitHub and Playwright MCP servers are enabled by default. GitHub also limits MCP tool calls made by Copilot code review to read-only operations.

Review comments now show when extra context was used

The generally available release also adds attribution to comments produced with Agent Skills or MCP context. When one of these sources contributes to a review comment, Copilot can indicate that additional context was involved.

That detail can make AI-assisted reviews easier to evaluate because developers can see when a recommendation depends on repository instructions or an external MCP source. Teams that already configured Skills or MCP during the public preview do not need to migrate their existing setup for the generally available release.

IMPORTANT: Agent Skills and MCP support in Copilot code review are generally available for Copilot Pro, Pro+, Business, and Enterprise users. MCP tool calls performed by code review are limited to read-only operations.{alertWarning}

Daisuki's Take: What This Means for Web Designers

The useful change here is not simply that Copilot can access more information. It is that code review can begin to reflect the rules and context of a particular web project instead of treating every repository as if it followed the same development standards.

For teams maintaining component libraries, templates, design systems, or larger frontend projects, that can make AI review more relevant. Repository-specific skills can provide recurring guidance, while MCP can connect reviews with documentation and operational context that would otherwise need to be checked separately.

We would still keep human review as the final control. Skills and external context can make Copilot more informed, but teams should verify that the instructions, MCP sources, and repository standards feeding the review remain accurate and appropriate for the current project.

{getCard} $type={custom} $title={What's New?} $info={Find more AI Web design news} $icon={}

Sources and Recommended Links

• Copilot code review: Agent skills and MCP now generally available | GitHub Official Changelog
• About agent skills | GitHub Copilot Official Documentation
• Configure MCP servers for your repository | GitHub Copilot Official Documentation