MCP server that saves Claude Code tokens by delegating bounded tasks to local or cloud LLMs. 93% token savings benchmarked. Works with LM Studio, Ollama, vLLM, DeepSeek, Groq, Cerebras.