Houtini LM
An MCP server that lets Claude Code hand off bounded coding tasks to a local or cheaper cloud model, cutting token spend while Claude reviews the returned work.
Houtini LM is an MCP server that lets Claude Code, or any MCP client, delegate bounded work, such as generating boilerplate, code review, commit messages, or format conversion, to another model rather than doing it with Claude's own reasoning and tool access. That other model can be a local model running in LM Studio, Ollama, or vLLM, OpenAI's models called directly, a model behind a LiteLLM router, one of OpenRouter's catalog, or a low-cost provider such as DeepSeek, while Claude continues to handle planning, multi-file changes, and judgment calls and checks whatever comes back.
It reads which real model sits behind each alias when pointed at a LiteLLM router, including that model's context window and output cap, so a mixed fleet of local and hosted models is sized correctly, and it adapts its request parameters automatically for models such as OpenAI's reasoning models that reject parameters other open models accept.
It is written in TypeScript, distributed as an MCP server package, and includes a Docker deployment guide.