GitHub Copilot agent mode with MCP support for JetBrains, Eclipse, and Xcode is now in public preview! You can enable GitHub Copilot with any GitHub account and try the autonomous coding experience today.

✨ What’s new

Agent mode

With the introduction of agent mode in these three IDEs, Copilot functions as an autonomous pair programmer that can perform multi-step coding tasks at your command, such as analyzing your codebase, proposing edits, and executing terminal commands using a defined set of tools. With agent mode, Copilot can even iterate on its own code, detect errors, and fix them automatically.

Each tool in the agent toolkit has specific capabilities that help Copilot complete these tasks. For example, Copilot can search the workspace, read file contents, execute terminal commands, retrieve compile or lint errors from the editor, and apply code changes via a speculative decoder endpoint. The set of available tools continues to grow as we experiment with new ones to further enhance the capabilities of Copilot in agent mode.

Here’s what a typical workflow looks like in agent mode after you enter your request:

  • Semantic understanding: Copilot interprets your request in Copilot Chat and analyzes the codebase to understand the task’s intent and requirements.

  • Plan proposal: Copilot leverages your code to build context around the request. Based on its understanding, it breaks the task into smaller steps and formulates a plan.

  • Plan execution: Copilot asks for confirmation to take the actions outlined in the previous bullet. Some examples of actions it might take include:

    • Multi-file manipulation: Creating or updating multiple files as required. (Demoed below in the JetBrains IDE).
      Agent mode for JetBrains, showing multi-files manipulation

    • Terminal execution: Executing terminal commands for the related tasks such as building the project and installing dependencies. (Demoed below in the Eclipse IDE).
      Agent mode for Eclipse, showing terminal execution

    • Error fixing: Fixing errors during the execution (Demoed below in Xcode).
      Agent mode for Xcode, showing error fixing

  • Task complete: After successfully implementing the changes, Copilot reviews its work, finalizes the task, and provides you with a summary of what changed.

MCP support

Model Context Protocol (MCP) is an open standard that enables AI models to interact with external tools and services through a unified interface. With Copilot supporting MCP in these three IDEs, agent mode gets extended capabilities from the tools contributed by MCP servers, so Copilot can assist across more workflows and tasks.

Set up your local MCP servers

You can define your servers in a configuration file (e.g., mcp.json) by doing the following:

  • In JetBrains IDEs: Click the GitHub Copilot icon -> Edit settings -> find the MCP Servers section.
  • In Eclipse: Click the GitHub Copilot icon -> Edit Preferences -> find the MCP Servers section
  • In Xcode: Open the GitHub Copilot for Xcode App -> Click the tool picker in Agent Mode -> Select Edit Config

Invoke tools directly in agent mode

Once configured, you can use natural language to call tools exposed by your MCP server. For example, you can ask Copilot to get the last modified time of the current files, and it will route the request through the appropriate tool. (Demoed below in Xcode).
Agent mode for JetBrains, MCP support

🛠 Try it out

Copilot agent mode is in preview and available today to all JetBrains,Eclipse, and Xcode users. You need to update the plugin or app to the latest version. To access Copilot agent mode for these IDEs, you also need a Copilot license.

💬 Share your feedback

Your feedback drives improvements. Let us know what you think using the in-product feedback option or share your thoughts in the following channels:

Disclaimer: The UI for features in public preview is subject to change.