Connect GitHub Copilot to JetBrains Air in Four Steps with ACP

GitHub Copilot recently became one of the native default agents in JetBrains AI Assistant. Now, thanks to ACP (Agent Client Protocol), the same GitHub Copilot agent can also run inside JetBrains Air.

ACP defines a standard way for an editor and an AI coding agent to communicate. Instead of building a separate integration for every editor, an agent can implement ACP once and connect to any compatible client. Air provides the client; GitHub Copilot provides the agent.

This guide skips the protocol deep dive and focuses on the practical part: connecting GitHub Copilot to Air in four steps.

Prerequisites

Before starting, make sure that:

• Air is installed and running.
• Node.js is installed, and npx is available in your terminal.
• A GitHub account.

Step 1: Add an ACP agent in Air

Open Air, click the agent selector at the bottom of the task view, and select Add ACP Agent....

Air will open its acp.json configuration file. This file defines the commands Air uses to start external ACP agents.

Step 2: Configure GitHub Copilot

Add the following configuration to acp.json:

{
    "agent_servers": {
        "GitHub Copilot npx": {
            "command": "npx",
            "args": [
                "@github/copilot-language-server",
                "--acp"
            ]
        }
    }
}

The important part is the --acp argument. It starts @github/copilot-language-server as an ACP agent, allowing Air to communicate with it as an ACP client.

You can rename GitHub Copilot npx if you prefer. The command and args values are what determine which agent is launched.

This setup uses GitHub Copilot as the product. The language server package is its ACP entry point; you are not using Copilot CLI as the user interface.

Step 3: Sign in to GitHub

Save acp.json, return to the agent selector, and choose the GitHub Copilot agent you just added.

The first time it starts, Air displays an Authentication required message. Click Sign in with GitHub and complete the authorization flow.

After authentication, GitHub Copilot is ready to work inside the current Air task.

Step 4: Choose a model and start coding

Open the GitHub Copilot model selector. You can leave it on Auto or choose one of the models available to your account.

Then give it a real coding task. For example:

Read this repository, fix the failing test, run the relevant test suite, and explain the change.

That is a better ACP test than a simple chat prompt. It exercises the full agent workflow: reading project context, editing files, running commands, and iterating on the result.

The result

GitHub Copilot is now running as an agent inside Air. The entire integration is described by one small JSON entry; no editor-specific GitHub Copilot plugin is required for this workflow.

The same pattern applies to other ACP-compatible agents. Their command and arguments will differ, but the integration model stays the same: add an agent server to acp.json, start it from Air, authenticate if necessary, and begin a task.

That is the practical power of ACP. Editors can support an ecosystem of agents, agents can reach more development environments, and developers can choose both sides independently.