This paper introduces BMPP (Blindly Meaningful Prompting Protocol), a protocol for safe and performant data exchange between distributed systems and Large Language Models (LLMs). BMPP inherits from Blindly Simple Protocol Language (BSPL) all the validation and verification features while adapting them for better performance when working with LLMs and agentic AI systems.

BMPP provides a formal grammar and a comprehensive toolchain that can generate production-ready code, with Rust as the reference implementation language. The protocol enables interoperable communication between heterogeneous agent systems while maintaining formal verification guarantees and optimizing for the structured generation capabilities of modern LLMs.