⚡ claude-gateway

Multi-provider gateway for Claude Code — route your Claude Code requests through OpenAI, Gemini, Ollama, Mistral, and more.

What is this?

A local proxy server that sits between Claude Code and any LLM provider. Claude Code sends Anthropic API requests → the gateway translates them to the active provider's format → streams responses back as Anthropic-format SSE.

Claude Code  ──→  claude-gateway (localhost:4747)  ──→  OpenAI / Gemini / Ollama / ...

Supported Providers

| Provider | Models | Auth | |---|---|---| | OpenAI | GPT-4o, GPT-5.x, o-series | OPENAI_API_KEY | | Google Gemini | Gemini 2.5 Pro, 2.0 Flash | GEMINI_API_KEY | | Ollama | Llama, Qwen, DeepSeek, Codestral (local) | None | | Mistral | Devstral, Codestral, Mistral Large | MISTRAL_API_KEY | | NVIDIA NIM | Nemotron, Llama, DeepSeek | NVIDIA_API_KEY | | MiniMax | MiniMax M2.5, M1 | MINIMAX_API_KEY | | GitHub Copilot | GPT-4o, Claude, o3-mini | GITHUB_TOKEN | | OpenAI Codex | GPT-5.x reasoning models | OPENAI_API_KEY |

Quick Start

1. Install

npm i -g claude-gateway

2. Set up a provider

# Set your OpenAI key
claude-gateway key openai sk-your-key-here

# Or use Ollama (no key needed, just have Ollama running)
claude-gateway switch ollama

3. Start the gateway

claude-gateway start

4. Configure Claude Code

Add to your ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:4747"
  },
  "mcpServers": {
    "gateway": {
      "command": "claude-gateway",
      "args": ["mcp"]
    }
  }
}

Now Claude Code routes through your chosen provider, and you can switch providers from within Claude Code using the MCP tools!

CLI Commands

claude-gateway              # Start the proxy server
claude-gateway start        # Start the proxy server
claude-gateway mcp          # Start MCP server (for Claude Code integration)
claude-gateway status       # Show current status
claude-gateway providers    # List all providers
claude-gateway switch <id>  # Switch active provider
claude-gateway key <id> <k> # Set provider API key
claude-gateway help         # Show help

MCP Tools (inside Claude Code)

Once configured as an MCP server, these tools are available inside Claude Code:

• list_providers — Show all providers and their status
• switch_provider — Switch to a different provider
• list_models — List models for a provider
• switch_model — Change the active model
• set_api_key — Configure a provider's API key
• gateway_status — Check gateway status

Configuration

Config is stored at ~/.claude-gateway/config.json:

{
  "activeProvider": "openai",
  "activeModel": "gpt-4o",
  "port": 4747,
  "providers": {
    "openai": {
      "apiKey": "sk-...",
      "defaultModel": "gpt-4o"
    },
    "gemini": {
      "apiKey": "AIza..."
    }
  }
}

You can also use environment variables (OPENAI_API_KEY, GEMINI_API_KEY, etc.) instead of storing keys in config.

How it works

1. Claude Code sends a POST /v1/messages request (Anthropic format)
2. The gateway translates the request to the active provider's API format
3. The provider processes the request and returns a response
4. The gateway translates the response back to Anthropic format
5. Claude Code receives a response it understands

For streaming requests, the gateway converts provider-specific SSE events into Anthropic's message_start → content_block_delta → message_stop event sequence.

License

MIT