Claude Proxy

Use any LLM provider with Claude Code. One command setup. Easy toggle on/off.

Quick Start

npm install -g claude-shadow
claude-shadow

The interactive wizard will ask you:

1. Which provider? (DeepSeek, OpenRouter, Ollama, Groq, etc.)
2. Your API key
3. Which model?
4. Start server + enable proxy? → Yes!

That's it. Claude Code now uses your chosen LLM.

Commands

| Command | Description | |---|---| | claude-shadow | Interactive setup wizard | | claude-shadow start | Start proxy server in background | | claude-shadow stop | Stop proxy server | | claude-shadow on | Point Claude Code to proxy | | claude-shadow off | Restore native Anthropic | | claude-shadow status | Show everything | | claude-shadow add | Add a custom provider | | claude-shadow restart | Restart the server |

Toggle Between Proxy and Native Claude

claude-shadow on     # Use DeepSeek/Gemini/etc.
claude-shadow off    # Back to native Anthropic

No manual settings.json editing. The toggle backs up your settings before modifying.

Manual Setup (Custom Port)

If you want to use a different port or set things up manually:

# Run the wizard, it asks which port you want
claude-shadow setup

# Or edit the config directly
cat ~/.claude-shadow/config.json

{
  "port": 9999,
  "apiKey": "sk-your-key",
  "provider": "deepseek",
  "model": "deepseek-chat",
  "apiBaseUrl": "https://api.deepseek.com/chat/completions"
}

Then:

claude-shadow start   # Start on port 9999
claude-shadow on      # Configure Claude Code
claude-shadow off     # Restore original settings

Built-in Providers

• DeepSeek — deepseek-chat, deepseek-reasoner
• OpenRouter — Claude, Gemini, GPT models via one API
• Ollama — Local models (no API key needed)
• Groq — Fast inference
• SiliconFlow — Chinese provider with DeepSeek/Qwen models
• Volcengine — ByteDance provider

Custom Providers

claude-shadow add

Enter name, API URL, key, and models. Works with any OpenAI-compatible API.

How It Works

Claude Code ──(Anthropic format)──▶ localhost:6666 ──(OpenAI format)──▶ DeepSeek/Gemini/etc.

claude-shadow runs a local server that:

1. Receives Anthropic-format requests from Claude Code
2. Converts them to OpenAI format
3. Forwards to your chosen provider
4. Converts the response back to Anthropic format
5. Returns it to Claude Code

Claude Code has no idea it's not talking to Anthropic.

Files

| Path | Purpose | |---|---| | ~/.claude-shadow/config.json | Proxy configuration | | ~/.claude-shadow/settings-backup.json | Original Claude Code settings backup | | ~/.claude-shadow/custom-providers.json | Your custom providers | | ~/.claude-shadow/proxy.pid | Server process ID |

Troubleshooting

Server won't start? Check if port is already in use:

lsof -i :6666

Claude Code still using Anthropic? Make sure proxy is enabled:

claude-shadow status
claude-shadow on

Switch back to native Claude:

claude-shadow off