Mcp Llm Gateway

This is a proxy layer that sits between Claude and any OpenAI-compatible LLM API. You point it at a downstream provider via DOWNSTREAM_URL, set a default model, and it exposes two MCP tools: list_models() to see what's available and complete() to send prompts with configurable temperature and max tokens. It also surfaces resources for the model list and current config. Reach for this when you want Claude to route completion requests to a different LLM provider without changing your client setup, or when you need to query multiple models through a single gateway. It handles API key passthrough and model discovery from endpoints like models.dev, so you can swap providers by changing environment variables.