Agent-to-Agent Communication (A2A)

Key points

• A2A = the 'language and channel' agents use to talk to each other.
• Messages are usually structured (clear sender, receiver, content).
• A 'handoff' is when one agent passes the whole task to another.
• Different from MCP: MCP connects agents to TOOLS; A2A connects agents to AGENTS.

The one-line definition

Agent-to-Agent (A2A) communication is how agents exchange messages, results, and tasks with each other in a structured, agreed-upon way, so a team of agents can cooperate without misunderstanding each other.

Note: A2A is the postal system between agents: clear sender, receiver, and a well-formatted message.

Three ways agents pass information

1) DIRECT HANDOFF 2) SHARED MESSAGE BUS ───────────────── ─────────────────────

Agent A Agent A Agent B │ "here, you take it" │ │ ▼ ▼ ▼ Agent B ┌──────────────────┐ (now owns the task) │ MESSAGE BUS │ │ (shared mailbox) │ └────────┬─────────┘ 3) BROADCAST ▼ ──────────── Agent C reads Agent A ──► everyone whatever it needs "status update for all"

What a structured message looks like

Agents should not just dump raw text at each other. They use structured messages, like a form with labelled fields, so the receiver always knows who sent it, what it is about, and what to do. Think of it like an envelope with a clear address, subject line, and contents, instead of a torn note slipped under a door.

• from / to: which agent sent it, which should read it.
• type: is this a request, a result, or an error?
• content: the actual data or task.

A tiny example of a structured A2A message

Here a researcher agent sends its findings to a writer agent. Notice the message is a clear dictionary with labelled fields, not a wall of text. Your coordinator code reads to and delivers it to the right agent.

# A simple structured message one agent sends to another
message = {
    "from": "researcher",
    "to": "writer",
    "type": "result",          # request | result | error
    "task_id": "report-42",
    "content": {
        "facts": ["EV batteries last 8-15 years",
                  "Recycling recovers ~95% of metals"],
    },
}

def deliver(message, agents):
    receiver = agents[message["to"]]   # find the target agent
    return receiver.handle(message)     # hand the message over

deliver(message, {"writer": writer_agent})

A2A vs MCP (a common point of confusion)

AGENT ◄────── A2A ──────► AGENT (talks to other AGENTS, shares results, hands off tasks) │ │ MCP ▼ ┌───────────┐ ┌───────────┐ ┌───────────┐ │ 🔧 TOOL │ │ 🗄️ DB │ │ 🌐 API │ └───────────┘ └───────────┘ └───────────┘ (MCP connects ONE agent to its TOOLS / data)

RULE OF THUMB: MCP = agent → tools A2A = agent → another agent

Standard protocols: Google's A2A and how it relates to MCP

People building agents kept inventing their own message formats, which made agents from different teams unable to talk. So standard protocols appeared. Google's A2A protocol is an open standard for how independent agents discover each other and exchange tasks and results in a common format, even if they were built by different companies.

Keep these two straight: MCP (Model Context Protocol) is for connecting an agent to its tools and data. A2A is for connecting an agent to other agents. They solve different problems and are often used together.

When do you need a real A2A setup?

A2A mistakes beginners make

Remember these lines

• A2A is how agents talk to OTHER agents; MCP is how an agent reaches its TOOLS.
• Always use structured messages: from, to, type, content.
• A handoff transfers ownership of the task to exactly one agent.
• Use a standard protocol (like Google's A2A) when agents from different systems must cooperate.

Key takeaways

• A2A communication is the structured way agents exchange messages, results, and tasks.
• Common channels: direct handoff, a shared message bus, and broadcast.
• Structured messages (from/to/type/content) prevent misunderstandings between agents.
• Google's A2A protocol is an open standard so agents from different systems can cooperate.
• MCP connects an agent to tools/data; A2A connects an agent to other agents.

Frequently Asked Questions

Related topics

• Goal Setting & Monitoring
• Learning & Adaptation
• Memory Management Pattern
• Multi-Agent Collaboration
• Prompt Chaining Pattern

Browse all Agentic AI Patterns topics →