Haohao-end/mcp-agent

Decouples LLM reasoning from tool execution through a stdio-based transport layer, enabling async multi-turn conversations with automatic tool schema conversion for OpenAI-compatible models. The RAG component combines LangChain document ingestion with ChromaDB vector storage and Maximal Marginal Relevance search, while weather and custom servers integrate via FastMCP's standardized function definitions. Built on asyncio for non-blocking I/O across client lifecycle management, external API calls, and concurrent tool invocations.