Copilot

Use your existing GitHub Copilot subscription as an in-editor agent.

If you already pay for GitHub Copilot, you can use it with NeoStack AI through Copilot CLI — no extra subscriptions needed. Copilot runs as an in-editor agent through ACP. There is no marketplace plugin path for Copilot — for driving the editor from a terminal, use Claude Code CLI or ChatGPT Codex CLI.

Requires NeoStack AI plugin 3.1.8 or newer.

Prerequisites

• A GitHub Copilot subscription (Individual, Business, or Enterprise)
• Node.js 22+ if installing via npm

Step 1 — install Copilot CLI

npm install -g @github/copilot

brew install copilot-cli

winget install GitHub.Copilot

curl -fsSL https://gh.io/copilot-install | bash

Verify with copilot --version.

Step 2 — authenticate

copilot login

Step 3 — use in Unreal Editor

1. Open Agent Chat (Tools → Agent Chat).
2. Select Copilot CLI from the agent dropdown.
3. Start chatting — the plugin spawns Copilot in ACP mode with access to all editor tools.

The plugin automatically writes the MCP server configuration to ~/.copilot/mcp-config.json on first launch.

Manual configuration (standalone)

To use Copilot CLI in a terminal outside the editor’s chat panel, connect it through the NeoStack bridge, a small stdio MCP binary that finds the running editor automatically.

Download the bridge:

On macOS/Linux, run chmod +x on it once.

Then run /mcp add in Copilot CLI, or add this to ~/.copilot/mcp-config.json:

{
  "mcpServers": {
    "unreal": {
      "type": "local",
      "command": "/path/to/neostack-bridge-macos-arm64",
      "tools": ["*"]
    }
  }
}

There are no ports, tokens, or URLs to configure. The bridge finds the running editor and follows restarts. If several editors are open, ask the agent to “list unreal projects” and “select” one, or set the NEOSTACK_PROJECT_DIR env var in the server config.

Plugin settings

Settings are in Project Settings → Plugins → NeoStack AI → Path Overrides.

Features

• GitHub Copilot’s agentic capabilities inside Unreal Editor
• Native ACP support — no separate adapter needed
• Streaming responses with tool call support
• Permission requests for tool execution
• Multiple model support

Troubleshooting