The agents panel provides an interface to delegate tasks to the Copilot coding agent from any page on github.com. Users can start tasks with simple natural-language prompts or reference issues and pull requests for context. The system can run multiple tasks in parallel and translates instructions into concrete action steps such as adding tests or UI features. Copilot has read access to repository data via the Model Context Protocol and can validate pages through a Playwright MCP Server, with support for custom MCP servers. The agent runs in the cloud, executes builds, tests, and linters in a secure environment, and produces draft pull requests while remaining active when users are offline.
The agents panel is designed for easy operation. Users can start new tasks with simple prompts, such as "Add integration tests for LoginController" or "Add a dark mode/light mode switcher." It is also possible to refer to specific GitHub issues or pull requests for context. For more complex scenarios, developers can perform multiple tasks in parallel. For example, unit tests for different files can be added at the same time. The system analyzes the natural language instructions and converts them into concrete action steps.
Thanks to the Model Context Protocol, Copilot has read access to repository data on GitHub and can view web pages for validation via the Playwright MCP Server. Custom MCP servers can also be linked to add specific functionality. Copilot works as a team member in the background In recent months, AI agents have taken on an increasingly prominent role in development workflows. Earlier this year, Microsoft turned GitHub Copilot into a fully-fledged AI agent that can reason and make decisions independently.
Collection
[
|
...
]