Houtini LM

Delegates bounded work from Claude Code to a local or cheaper cloud LLM - code review, test stubs, commit messages, format conversion - so those tokens stay off your Claude bill. Auto-profiles the models your server has loaded via HuggingFace and routes each task to the best fit. Works with LM Studio, Ollama, vLLM, DeepSeek, Groq, Cerebras and OpenRouter. Install: `npx -y @houtini/lm`. Apache-2.0.

Project

Browse this MCP server in the TensorBlock MCP Index.