Copilot API Proxy

Turn GitHub Copilot into an OpenAI/Anthropic/Gemini API compatible server. Usable with Claude Code, Codex, OpenCode, and more.

Installation

Run directly (no install)

bunx @pointgroup/copilot-api start -M
npx @pointgroup/copilot-api start -M

Global install

npm install -g @pointgroup/copilot-api

From source

git clone https://github.com/pointgroup-labs/copilot-api.git
cd copilot-api
bun install
bun run build
bun link

Authentication

First-time setup (interactive)

copilot-api auth

Opens the GitHub device flow — enter the code at github.com to authorize. Use --show-token to print the token for reuse:

copilot-api auth --show-token

Using a saved token

copilot-api start -g YOUR_GITHUB_TOKEN

Quick Start

# Start with defaults (port 4141)
copilot-api start

# Native Claude Messages API (recommended for Claude models)
copilot-api start -M

# With a saved GitHub token + native messages
copilot-api start -g YOUR_GITHUB_TOKEN -M

# Custom port
copilot-api start -p 8080

# Business account
copilot-api start -a business

# Enterprise account
copilot-api start -a enterprise

CLI Reference

Commands

| Command | Description | |:--------|:------------| | copilot-api auth | Run GitHub auth flow | | copilot-api start | Start the API server | | copilot-api check-usage | Show current Copilot quota | | copilot-api debug | Print debug information |

Start Options

| Flag | Short | Default | Description | |:-----|:------|:--------|:------------| | --port | -p | 4141 | Port to listen on | | --verbose | -v | | Enable verbose logging | | --account-type | -a | individual | Account type (individual, business, enterprise) | | --github-token | -g | | Provide GitHub token directly | | --native-messages | -M | | Use Copilot's native /v1/messages for Claude models | | --force-agent | -F | | Smart agent: auto-switch to agent mode when over quota | | --claude-code | -c | | Generate a Claude Code launch command | | --rate-limit | -r | | Rate limit in seconds between requests | | --wait | -w | | Wait instead of error on rate limit | | --manual | | | Manual request approval | | --show-token | | | Show tokens on fetch/refresh | | --proxy-env | | | Initialize proxy from environment variables |

API Endpoints

| Endpoint | Format | Description | |:---------|:-------|:------------| | POST /v1/messages | Anthropic | Claude Messages API | | POST /v1/messages/count_tokens | Anthropic | Token counting | | POST /v1/chat/completions | OpenAI | Chat Completions API | | POST /v1beta/models/{model}:generateContent | Gemini | Gemini API | | POST /v1/responses | OpenAI | Responses API | | GET /v1/models | OpenAI | List models with capabilities |

Usage with Claude Code

# Generate the launch command automatically
copilot-api start -M -c

# Or configure manually
ANTHROPIC_BASE_URL=http://localhost:4141 \
ANTHROPIC_API_KEY=anything \
claude

Configuration

Config file location: ~/.local/share/copilot-api/config.json

{
  "extraPrompts": {},
  "smallModel": "gpt-5-mini",
  "compactUseSmallModel": true,
  "useFunctionApplyPatch": true,
  "modelReasoningEfforts": {
    "gpt-5-mini": "low",
    "claude-opus-4.6": "xhigh"
  },
  "webSearchModel": "claude-sonnet-4"
}

| Option | Type | Default | Description | |:-------|:-----|:--------|:------------| | extraPrompts | Record<string, string> | | Model-specific system prompt additions | | smallModel | string | gpt-5-mini | Model for warmup/compact requests | | compactUseSmallModel | boolean | true | Use small model for compact/summarization requests | | useFunctionApplyPatch | boolean | true | Convert custom apply_patch to function type | | modelReasoningEfforts | Record<string, string> | | Per-model reasoning effort levels | | webSearchModel | string | (disabled) | Model to use when web search tools are detected (e.g. claude-sonnet-4). Saves premium quota on search-heavy requests |

Development

bun run dev          # Dev server with watch
bun run dev -- -M    # Dev with native messages
bun test             # Run all tests
bun run typecheck    # Type check
bun run lint:all     # Lint

License

MIT