MPOP (Multi-LLM Peer Orchestration Protocol) defines a standardized framework for enabling multiple Large Language Models to participate in shared conversations as peers. Abstract Unlike traditional single-model interactions, MPOP establishes conventions for context sharing, inter-model addressing, turn arbitration, and human authority preservation. This specification is transport-agnostic and implementation-independent, allowing adoption across web applications, mobile clients, CLI tools, and embedded systems. Key Innovations Context Sharing: All participating models receive the complete conversation history including contributions from other models @Mention Addressing Protocol: Standardized syntax for direct inter-model communication Dynamic Adapter Profiles (DAP): Traffic control mechanism that adjusts participation probabilities without editorial control Four-Layer Architecture: L0 (Adapter) → L1 (Facilitator) → L2 (Governance) → User Layer Dual Operating Modes: Collaborative v2.5 ("What am I missing?") and Adversarial v3.1 ("Prove it.") Relationship to MCP MPOP complements the Model Context Protocol (MCP) which standardizes tool integration. Where MCP defines how models access external tools and data sources, MPOP defines how models communicate with each other. Conformance Levels The specification defines four conformance levels: Basic (Level 1), Standard (Level 2), Full (Level 3), and Extended (Level 4), allowing implementations to claim partial compliance. Document ID: MPOP-2026-01

Protocol specification developed through iterative design and implementation experience with multi-LLM orchestration systems. Informed by production deployment of AI Peer Chat platform.

This is the initial release of the MPOP specification. The protocol draws inspiration from the Model Context Protocol (MCP) and extends the paradigm to model-to-model communication.