Agent to agent protocol: how A2A and MCP fit together
Two open protocols come up whenever people build with AI agents. MCP connects an agent to tools and data. A2A connects an agent to other agents. This site explains both, compares them, and lets you watch two simulated agents talk to each other message by message.
The protocol map
The A2A documentation uses a simple picture: MCP is vertical, because it deepens what a single agent can do. A2A is horizontal, because it reaches across the boundary between agents.
Short version
- MCP (Model Context Protocol): an AI application talks to servers that expose tools, resources and prompts.
- A2A (Agent2Agent): a client agent discovers a remote agent through its Agent Card, sends it a task, follows status updates, and receives artifacts.
- They are complementary. A real system often uses both.
Where to go next
- A2A vs MCPDifference table, when to use which, and a five question quiz.
- Live demoTwo agents, every message visible, runs in your browser.
- What is A2AAgent Card, tasks, messages, artifacts, opacity.
- Agent Card validatorPaste an A2A Agent Card and check it against spec 1.0.0.
- What is MCPHosts, clients, servers, and the three primitives.
FAQ
- What is an agent to agent protocol?
- It is a shared message format and set of rules that lets one AI agent find another agent, send it work, and get status and results back, even when the two agents come from different vendors. A2A (Agent2Agent) is the main open protocol for this.
- Is A2A a replacement for MCP?
- No. The official A2A documentation describes the two as complementary. MCP connects an agent to tools and data. A2A connects an agent to other agents.
- Can I see AI agents talking to each other?
- Yes. The demo page runs two simulated agents in your browser and shows every A2A-shaped message in a transcript. It is a simulation, not a production A2A server.