Copilot Proxy Go

Use GitHub Copilot with Claude Code, Codex CLI, Cursor, OpenCode, and any OpenAI/Anthropic-compatible client.

copilot-proxy-go is a local Go proxy that turns your GitHub Copilot subscription into OpenAI-compatible and Anthropic-compatible API endpoints, with first-class setup for Claude Code and Codex CLI.

Fork notice: This repository is based on a fork of tonghaoch/copilot-proxy-go and includes additional improvements and bug fixes.

Works Great With

Tool	One-command setup	API
Claude Code	./copilot-proxy-go start --claude-code	Anthropic Messages API
Codex CLI	./copilot-proxy-go start --codex	OpenAI Responses API

Highlights

• Claude Code ready — interactive model selection and environment generation with --claude-code
• Codex CLI ready — generates a Responses API configuration with --codex
• VPN/proxy friendly — use --proxy-env to route GitHub/Copilot requests through http_proxy, https_proxy, or all_proxy
• OpenAI + Anthropic compatible — supports Chat Completions, Responses, Messages, and Embeddings
• Copilot-backed — uses your GitHub Copilot account as the backend model provider
• Token management — GitHub OAuth device-code flow with automatic Copilot token refresh

Quick Start

1. Build or install

Build from source:

go build -o copilot-proxy-go .

Or install the latest version directly from GitHub:

go install github.com/emilioforrer/copilot-proxy-go@latest

If you install with go install, use copilot-proxy-go instead of ./copilot-proxy-go in the commands below.

2. Authenticate

./copilot-proxy-go auth

This opens a GitHub device-code flow in your browser. The token is saved to ~/.local/share/copilot-proxy-go/github_token.

3. Start for your client

Claude Code:

./copilot-proxy-go start --claude-code

Codex CLI:

./copilot-proxy-go start --codex

Generic OpenAI/Anthropic-compatible server:

./copilot-proxy-go start

The proxy runs on http://localhost:4141. OpenAI-compatible clients should use http://localhost:4141/v1 as the base URL.

Behind a VPN or proxy

If GitHub or Copilot requires a proxy in your network, export your proxy variables and start with --proxy-env:

export https_proxy=http://127.0.0.1:7890 \
  http_proxy=http://127.0.0.1:7890 \
  all_proxy=socks5://127.0.0.1:7890

./copilot-proxy-go start --proxy-env

Background Service

Install and manage the proxy as a per-user background service:

copilot-proxy-go service install
copilot-proxy-go service start
copilot-proxy-go service status

service install registers the service but does not start it. The service runs as the installing user, uses that user's existing Copilot credentials and configuration, and does not require sudo. Linux uses a systemd user unit; macOS uses a launchd user agent.

Other service operations:

copilot-proxy-go service stop
copilot-proxy-go service restart
copilot-proxy-go service uninstall

The service normally starts when the user logs in. On headless Linux workers that must start before login, enable lingering once:

sudo loginctl enable-linger "$(whoami)"

Features

• Multi-API support — Chat Completions, Messages (Anthropic), Responses, Embeddings
• Automatic translation — routes Anthropic requests to the best available backend (native Messages API → Responses API → Chat Completions)
• Extended thinking — full support for Claude thinking/reasoning blocks with interleaved thinking protocol
• Streaming — SSE streaming with proper event translation across all API formats
• Quota optimization — auto-routes compact/warmup requests to smaller models to save premium quota
• Local dashboard — view usage, request metrics, session state, and quota information in your browser
• Background service — install and manage the proxy as a per-user systemd or launchd service.

Dashboard

Usage with Claude Code

./copilot-proxy-go start --claude-code

This interactively selects models and generates the environment variables for Claude Code. Or set them manually:

export ANTHROPIC_BASE_URL=http://localhost:4141
export ANTHROPIC_AUTH_TOKEN=copilot-proxy
export ANTHROPIC_MODEL=claude-sonnet-4
export ANTHROPIC_SMALL_FAST_MODEL=gpt-5-mini
claude

Usage with Codex CLI

./copilot-proxy-go start --codex

This interactively selects a Responses-capable model and generates a Codex command. Or set it manually:

export CODEX_API_KEY=copilot-proxy
export NO_PROXY=localhost,127.0.0.1,::1
export no_proxy=localhost,127.0.0.1,::1
codex \
  -c 'model="gpt-5.3-codex"' \
  -c 'model_provider="copilot-proxy"' \
  -c 'model_providers.copilot-proxy.name="Copilot Proxy"' \
  -c 'model_providers.copilot-proxy.base_url="http://127.0.0.1:4141/v1"' \
  -c 'model_providers.copilot-proxy.env_key="CODEX_API_KEY"' \
  -c 'model_providers.copilot-proxy.wire_api="responses"'

API Endpoints

Endpoint	Method	Description
/v1/messages	POST	Anthropic Messages API
/v1/messages/count_tokens	POST	Token counting
/chat/completions	POST	OpenAI Chat Completions
/v1/chat/completions	POST	OpenAI Chat Completions
/responses	POST	OpenAI Responses API
/v1/responses	POST	OpenAI Responses API
/embeddings	POST	Embeddings
/models	GET	List available models
/v1/models	GET	List available models
/dashboard	GET	Usage dashboard (web UI)

CLI Reference

start — Run the proxy server

copilot-proxy-go start [flags]

Flags:
  -a, --account-type string   individual, business, or enterprise (default "individual")
  -c, --claude-code           interactive model selection for Claude Code
      --codex                 interactive model selection for Codex CLI
  -g, --github-token string   GitHub OAuth token (skips device code flow)
      --host string           host/IP to bind to, use 0.0.0.0 for all interfaces (default "127.0.0.1")
  -p, --port int              port to listen on (default 4141)
      --proxy-env             enable HTTP proxy from env vars (http_proxy/https_proxy/all_proxy)
      --show-token            print tokens to console
  -v, --verbose               enable verbose/debug logging

auth — Authenticate with GitHub

copilot-proxy-go auth [flags]

Flags:
      --force        force re-authentication
      --show-token   print token to console

License

MIT