[0] https://modelcontextprotocol.io/specification/2025-06-18/ser...
> Wouldn't LLM need to directly output the correct machine code for that to be true?