Based on GitHub / Microsoft’s official docs, checked October 30, 2026

GitHub Copilot agent mode: how to use it in your IDE

In agent mode, GitHub Copilot takes a high-level task, decides which files to change, makes the edits and runs commands, and keeps iterating until the task is done. It works inside your editor: you review the changes, and by default you approve commands before they run. GitHub lists agent mode in Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode and Eclipse. This guide covers switching to it, steering it, picking a model and adding tools, based on GitHub’s documentation checked October 30, 2026. For the agent that runs in the cloud and opens pull requests, see our Copilot coding agent guide.

1. Open Copilot Chat
In VS Code, select Open Chat from the Copilot Chat menu. In Visual Studio (17.14 or later), choose View, then GitHub Copilot Chat. In JetBrains IDEs, choose Copilot in the menu bar, then Open GitHub Copilot Chat.
2. Select Agent
In VS Code and Visual Studio, pick Agent from the dropdown at the bottom of the chat; in JetBrains, click the Agent tab at the top. If Agent is missing, your enterprise or organization administrator may have disabled agent mode for your IDE.
3. Describe the task
Submit a prompt. Copilot streams edits into the editor, updates the working set and runs or suggests terminal commands as needed.
4. Approve commands and review the edits
Confirm or reject each terminal command it proposes unless it has been set to run automatically, and undo any streamed edit you don’t want. Copilot iterates on errors until the original task is done.

When to use agent mode

GitHub suggests agent mode when a task is complex and involves several steps, iterations and error handling, when you want Copilot to work out the steps itself, or when the task needs an external application such as an MCP server.

For a large or unclear task, draft an implementation plan first in plan mode, then switch to agent mode to carry it out.

Steering it while it works

Agent mode is interactive. You can send a follow-up prompt to redirect it before it finishes, reject commands, and undo edits as they appear.

To hand a self-contained subtask to a separate agent with its own context, VS Code supports subagents.

Models, custom agents and usage

Before you submit a task you can change the AI model, and in VS Code and JetBrains you can select a custom agent for a specific kind of work.

GitHub notes that each prompt you enter in agent mode consumes GitHub AI Credits; see our Copilot pricing page for plan details.

Extending it with MCP servers

Much of what agent mode can do comes from the tools it can call. Add Model Context Protocol servers to give it tools for external systems, or the GitHub MCP Server to work with GitHub from your editor.

Tips

Keep command approvals on until you trust a workflow; only allow automatic runs for commands you would run yourself without looking.
Start big changes in plan mode so you can correct the approach before files change.
Agent mode works on your local files, so commit before a large task and review the diff afterwards.