Introduction
Getting started with the Multi-Agent Coordination Protocol
The Multi-Agent Coordination Protocol (MACP) is an open protocol for coordinating autonomous AI agents. It provides a structural coordination kernel — deterministic, secure, and observable — for ecosystems where multiple capable agents must produce binding outcomes.
Why MACP?
As intelligence becomes abundant, coherence becomes the bottleneck. MACP provides the missing coordination kernel:
- Structured sessions with full lifecycle management
- Five coordination modes — Decision, Proposal, Task, Handoff, Quorum
- Deterministic execution with replay integrity
- Security first — capability-based auth and transport security
- Observable — built-in tracing, metrics, and structured logs
Quick Overview
Architecture
MACP defines three core concepts:
- Runtime — The coordination engine that manages sessions, enforces mode rules, and provides transport
- Agents — Autonomous participants that join sessions and exchange messages
- Sessions — Bounded coordination contexts with a specific mode, lifecycle, and outcome
Coordination Modes
Each mode defines how agents interact within a session:
| Mode | Purpose |
|---|---|
| Decision | Proposals, evaluations, votes, and a single binding outcome |
| Proposal | Proposal and counterproposal negotiation |
| Task | Bounded task delegation |
| Handoff | Responsibility transfer between participants |
| Quorum | Threshold-based approval and rejection |
Next Steps
- Architecture — Understand the protocol structure
- Coordination Modes — Deep dive into the five modes
- Examples — See MACP in action
- Specification — Read the full RFC set