AgentProtocol.ai
Protocol selectorCompare protocols
//how agents communicate

Agent communication protocol

An agent communication protocol defines how AI agents exchange messages, tasks and context in a structured, predictable way. This guide breaks agent communication into layers and shows where each modern protocol fits.

Concept

What is an agent communication protocol?

An agent communication protocol is a set of rules governing how agents send and receive information. It specifies the message format, the sequence of exchanges, and what each side can expect — so components that have never met can still cooperate.

The idea has deep roots. Earlier multi-agent research produced agent communication languages such as FIPA-ACL and KQML. Today's protocols are more pragmatic, built on web standards like HTTP, JSON and JSON-RPC, and focused on connecting LLM-based agents to tools and to each other.

The layers of agent communication

It helps to separate agent communication into three layers. Most confusion comes from mixing them.

  • Agent-to-tool — an agent calling functions, APIs or data sources. Standardised by MCP.
  • Agent-to-agent — independent agents delegating tasks. Standardised by A2A and described generally on our agent-to-agent page.
  • Client-to-agent — an application driving an agent through an API. Standardised by the Agent Protocol REST spec.
Why layers matter
A single system often uses all three at once. Naming the layer you're working at tells you which protocol to reach for and avoids the trap of treating complementary standards as competitors.

What a good protocol provides

  • A shared message format so both sides parse the same structure.
  • Capability discovery so a caller can learn what the other side can do at runtime.
  • A task lifecycle so long-running work can be started, tracked and completed.
  • Error semantics so failures are handled consistently rather than ad hoc.

How the modern protocols relate

MCP, A2A and Agent Protocol are all agent communication protocols, each covering a different layer. For a direct comparison see MCP vs A2A vs Agent Protocol. For the human-language framing of the same topic — “AI agent communication” — see the AI agent communication protocol guide.

//questions

Frequently asked questions

Is 'agent communication protocol' a specific standard?

It's mainly a general term for any protocol that governs how agents communicate. Several specific standards fall under it — notably MCP, A2A and the Agent Protocol REST spec — each covering a different communication layer.

How is this different from an API?

An API is one system's specific interface. A protocol is a shared convention many systems implement, so that any compliant component can interoperate. Some agent protocols (like Agent Protocol) are themselves defined as REST APIs.

What about older agent communication languages?

Standards like FIPA-ACL and KQML predate LLM agents and shaped the field, but today's widely used protocols are built on modern web foundations (HTTP, JSON, JSON-RPC) and focus on practical LLM-agent scenarios.

// AgentProtocol is independent and not the official source for any protocol named here.