@aistack/claude-code-proxy

A high-performance proxy server that translates Anthropic Claude API requests to LiteLLM format, enabling seamless integration with multiple LLM providers including OpenAI, Google Gemini, OpenRouter, and Anthropic.

Features

• 🚀 Multi-Provider Support: Seamlessly switch between OpenAI, Gemini, OpenRouter, and Anthropic models
• 🔄 Automatic Format Translation: Converts Anthropic API format to LiteLLM format and back
• 🎯 Smart Model Mapping: Intelligently maps Claude models to equivalent models from other providers
• 📡 Streaming Support: Full support for Server-Sent Events (SSE) streaming responses
• 🛡️ Production Ready: Comprehensive error handling, logging, and graceful shutdown
• ⚡ High Performance: Built with Go for optimal performance and low resource usage
• 🔧 Configurable: Flexible configuration via environment variables

Installation

Install globally using npm:

npm install -g @aistack/claude-code-proxy

Quick Start

1. Initialize configuration:

ccproxy init

This will prompt you to create a configuration file at either:

• User level: $HOME/.ccproxy/settings.json
• Project level: ./.ccproxy/settings.json

2. Edit the configuration file:

{
  "server": {
    "port": 8082,
    "host": "0.0.0.0"
  },
  "providers": {
    "openai": {
      "apiKey": "sk-...",
      "baseURL": "https://api.openai.com/v1",
      "responseStyle": "openai"
    },
    "anthropic": {
      "apiKey": "sk-ant-...",
      "baseURL": "https://api.anthropic.com",
      "responseStyle": "anthropic"
    },
    "gemini": {
      "apiKey": "...",
      "baseURL": "https://generativelanguage.googleapis.com",
      "responseStyle": "openai"
    },
    "openrouter": {
      "apiKey": "sk-or-v1-...",
      "baseURL": "https://openrouter.ai/api/v1",
      "responseStyle": "openai"
    }
  },
  "models": {
    "bigModel": "gpt-4@openai",
    "smallModel": "gpt-4-mini@openai"
  },
  "logging": {
    "level": "info",
    "format": "json"
  }
}

3. Start the proxy:

# Start proxy server and launch claude CLI (default behavior)
ccproxy

# Or start only the proxy server
ccproxy server

The server will start on http://localhost:8082 (or your configured port).

Commands

Available Commands

• ccproxy (default) - Start proxy server (if needed) and launch claude CLI
• ccproxy server - Start only the proxy server
• ccproxy init - Initialize configuration file interactively
  • --scope user - Create user-level configuration
  • --scope project - Create project-level configuration
• ccproxy help - Show help message
• ccproxy version - Show version information

Global Options

• --debug - Enable debug logging (applies to all commands)
• --port - Override server port (server command only)

Examples

# Start server and launch claude CLI (default behavior)
ccproxy

# Start with debug logging enabled
ccproxy --debug

# Start only the proxy server
ccproxy server

# Start server on custom port
ccproxy server --port 8080

# Start server with debug logging
ccproxy server --debug

# Initialize configuration interactively
ccproxy init

# Create user-level configuration
ccproxy init --scope user

# Create project-level configuration
ccproxy init --scope project

Usage

Configuration

Configuration Files

The proxy loads configuration from multiple sources in priority order:

1. Command Line Arguments (highest priority)
   • --port: Override server port
2. JSON Configuration Files:
   • /etc/ccproxy/managed-settings.json (managed/system-wide)
   • $CWD/.ccproxy/settings.json (project-level)
   • $HOME/.ccproxy/settings.json (user-level)

Model Format

Models are specified in the format modelname@provider:

• gpt-4@openai
• claude-3-opus@anthropic
• gemini-pro@gemini
• gpt-4@openrouter

Response Styles

Providers can have different response styles:

• "responseStyle": "anthropic" - Native Anthropic format (no transformation)
• "responseStyle": "openai" - OpenAI format (requires transformation)

This allows the proxy to skip unnecessary transformations for providers that natively support Anthropic's API format.

Model Mapping

The proxy automatically maps Claude models to configured models:

Configuration Options

JSON Configuration

Default Behavior

When you run ccproxy without arguments, it will:

1. Check if the proxy server is already running
2. Start the server if it's not running
3. Launch the claude CLI with ANTHROPIC_BASE_URL set to the proxy

This means you can use ccproxy as a drop-in replacement for the claude command while automatically getting multi-provider support.

Adding Custom Providers

You can add custom providers with native Anthropic support:

{
  "providers": {
    "custom-llm": {
      "baseURL": "https://your-llm-api.com",
      "apiKey": "your-api-key",
      "responseStyle": "anthropic"
    }
  }
}

Then use it with: custom-model@custom-llm

Platform Support

The CLI is available for the following platforms:

• macOS (Intel & Apple Silicon)
• Linux (x64 & ARM64)
• Windows (x64)

Troubleshooting

Binary Not Found

If you see "Binary not found" error, it means your platform is not supported. Please open an issue with your platform details.

Permission Denied

On Unix systems, if you get a permission error:

sudo npm install -g @aistack/claude-code-proxy

API Key Issues

Ensure your API keys are correctly set in your settings.json file and that they have the necessary permissions.

Development

For development or contributing, visit the GitHub repository.

License

MIT License - see the LICENSE file for details.

Support

• 📚 Documentation
• 🐛 Issue Tracker
• 💬 Discussions

Changelog

See CHANGELOG.md for release history.