Cursor OpenCode Auth

An OpenAI-compatible proxy server that lets you use Cursor's AI models (composer-1, claude-4.5-sonnet, gpt-5.2-codex, etc.) in OpenCode or any OpenAI-compatible client.

Requirements

• Node.js 18+
• macOS (uses Keychain for token storage)
• Cursor CLI installed and logged in - Install here (the proxy uses your Cursor auth token)

Quick Start

Step 1: Start the proxy server (run in a separate terminal)

# Install globally and run
npm install -g cursor-opencode-auth
cursor-proxy

# Or run directly with npx
npx cursor-opencode-auth

# Or clone and run
git clone https://github.com/R44VC0RP/cursor-opencode-auth
cd cursor-opencode-auth
node proxy-server.mjs

Step 2: Keep the proxy running while you use OpenCode

The server runs on http://localhost:4141 by default.

Usage with OpenCode

Add this to your opencode.json:

{
  "provider": {
    "cursor": {
      "name": "Cursor (Proxy)",
      "api": "http://localhost:4141/v1",
      "models": {
        "composer-1": {
          "name": "Composer 1",
          "limit": { "context": 200000, "output": 32000 }
        },
        "claude-4.5-sonnet": {
          "name": "Claude 4.5 Sonnet",
          "limit": { "context": 200000, "output": 16000 }
        },
        "gpt-5.2-codex": {
          "name": "GPT 5.2 Codex",
          "limit": { "context": 128000, "output": 16000 }
        }
      }
    }
  }
}

Then select cursor/composer-1 (or other model) in OpenCode.

Usage with curl

# List available models
curl http://localhost:4141/v1/models

# Chat completion
curl http://localhost:4141/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "composer-1",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

# Streaming
curl http://localhost:4141/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "composer-1",
    "messages": [{"role": "user", "content": "Hello!"}],
    "stream": true
  }'

Available Models

The proxy exposes all models available in your Cursor subscription:

• composer-1 - Cursor's flagship model
• gpt-5.2-codex - GPT 5.2 optimized for code
• claude-4.5-sonnet - Claude 4.5 Sonnet
• claude-4-opus - Claude 4 Opus
• gemini-2.5-pro - Gemini 2.5 Pro
• grok-3 - Grok 3
• o4 - O4 reasoning model
• And more...

How It Works

1. The proxy extracts your Cursor auth token from macOS Keychain
2. Incoming OpenAI-format requests are translated to Cursor's Connect-RPC/protobuf format
3. Requests are sent to Cursor's API (agentn.api5.cursor.sh)
4. Protobuf responses are parsed and converted back to OpenAI format

Configuration

# Custom port
node proxy-server.mjs 8080

# Enable debug logging
DEBUG=1 node proxy-server.mjs

Troubleshooting

"Could not get token from keychain"

Make sure you have the Cursor CLI installed and are logged in. The token is stored at:

security find-generic-password -s "cursor-access-token" -w

Empty responses

Check the proxy logs for errors. The proxy includes debug output showing:

• Request details (model, message count)
• Cursor API status
• Response size and extracted text

License

MIT