What is A2A? Agent-to-Agent Protocol Explained

Bottom line: A2A is a protocol that lets AI agents talk to each other. It complements MCP: MCP connects agents to tools, while A2A connects agents to agents.

The problem A2A solves

As AI agents multiply inside companies, they need a way to collaborate. One agent might specialize in research, another in code, and another in scheduling. Without a common protocol, each pair of agents needs a custom integration.

A2A standardizes how agents discover each other, advertise their capabilities, negotiate tasks, and return results. An agent can ask another agent, "Can you handle this?" and get a structured response without human intervention.

A2A vs MCP in one sentence

How A2A works

A2A uses a few core concepts:

• Agent Card: a JSON manifest describing what an agent can do, its endpoint, and authentication requirements.
• Task: a unit of work sent from one agent to another.
• Message: the request and response payloads exchanged during a task.
• Skill: a capability advertised in the Agent Card, similar to an MCP tool but at the agent level.

When Agent A wants help, it fetches Agent B's Agent Card, picks a skill, sends a task, and receives updates or a final result. The protocol supports streaming, push notifications, and structured output.

Real-world use cases

• Research team: a planner agent delegates web search, summarization, and fact-checking to specialist agents.
• Software team: a product agent hands off a feature request to an architecture agent, which hands off implementation to a coding agent.
• Enterprise workflow: a CRM agent asks a billing agent to generate an invoice after a deal closes.

When to use A2A

Use A2A when agents are autonomous, run in different environments, and need to collaborate on complex workflows. If you only need an agent to call a tool, MCP is simpler and more mature. Many systems will use both: MCP for tool access and A2A for agent coordination.

Published 2026-06-12