Copilot Claude Proxy

⚠️ Disclaimer: This is an unofficial, experimental tool. It interfaces with GitHub Copilot endpoints in ways that may not be officially supported for third-party clients. Users are responsible for complying with GitHub's and Anthropic's Terms of Service. Use at your own risk. The author is not responsible for blocked accounts or ToS violations.

A local compatibility proxy that accepts Anthropic-style POST /v1/messages calls and forwards them to the GitHub Copilot OpenAI-compatible chat endpoint. Ideal for using Claude Code with your existing Copilot models.

This is intended for personal, authorized use only.

🚀 Quick Start (Zero Install)

You can run this proxy directly using npx without needing to clone the repository or manually install dependencies.

1. Open your terminal and run:

npx @ace-ify/copilot-claude-proxy

2. The server will start on http://127.0.0.1:2169 by default.

3. Open the Admin UI at http://127.0.0.1:2169/admin in your browser.

4. Click Connect Copilot to authenticate via the GitHub OAuth device flow. This securely generates a token used to forward requests to GitHub models.

5. Select your preferred model and click Enable Proxy. This will automatically patch your Claude settings.

🛠️ Local Setup (For Developers)

If you prefer to run it locally from source:

1. Clone the repository and install dependencies:

git clone https://github.com/ace-ify/copilot-claude-proxy.git
cd copilot-claude-proxy
npm install

2. (Optional) Create an .env file:

copy .env.example .env

3. Run the server:

npm start

⚙️ Configuration (Environment Variables)

The proxy comes with sensible defaults, but you can override them using environment variables.

| Variable | Default | Purpose | | :--- | :--- | :--- | | PORT | 2169 | Local listen port. | | COPILOT_ACCESS_TOKEN | null | Your GitHub Copilot access token. If you use the Web UI to login, you don't need to set this. | | GITHUB_TOKEN | null | Fallback if COPILOT_ACCESS_TOKEN is unset. | | DEFAULT_MODEL | gpt-5.3-codex | Fallback model if the incoming request omits a model. | | ADMIN_MODEL_ALLOWLIST | multiple | Comma-separated model IDs visible in the Admin UI dropdown. |

Note: For a full list of advanced tuning variables (endpoint overrides, caching TTLs, beta headers), please check src/server.js.

🧠 How it Works & Routing

This proxy acts as a bridge between your local tools (like Claude Code) and GitHub Copilot's upstream APIs.

1. Model Discovery: The proxy checks Copilot /models and inspects each model's supported_endpoints.
2. Anthropic Native: If a model supports /v1/messages natively, the proxy forwards directly to Copilot /v1/messages.
3. Responses API: If a model supports /responses, the proxy uses /responses (supporting both non-streaming and streaming).
4. OpenAI Fallback: Otherwise, it falls back to Copilot /chat/completions and performs Anthropic <-> OpenAI mapping on the fly.

Endpoints

• POST /v1/messages: Anthropic-like messages payload mapped to Copilot.
• GET /healthz: Health check endpoint.
• /admin/*: Internal endpoints for the dashboard (status, trace, live metrics, auth).

🔒 Security Notes

• Keep this local only (127.0.0.1) unless you intentionally add auth and network controls.
• Do not share your token.
• Copilot upstream model availability depends on your GitHub Copilot entitlement and account policy.
• Upstream schema or policy changes may require adapter updates.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.