## Definition
The **Agent-to-Agent (A2A) protocol** is an open standard for agents — possibly built by different vendors or on different frameworks — to discover one another, delegate tasks, and exchange results. Where the [[Model Context Protocol]] connects an agent *down* to tools and data (vertical), A2A connects an agent *across* to **other agents** (horizontal).
## What it standardizes
- **Discovery** — an "agent card" advertising an agent's identity, skills, and endpoint.
- **Delegation** — send a task, stream its progress, and receive the result over a common message format.
- **Interoperability** — agents built on different stacks cooperate without a bespoke integration for each pair.
## MCP vs A2A
- [[Model Context Protocol|MCP]]: agent ↔ tools / data.
- A2A: agent ↔ agent.
They are complementary halves of an agent's connectivity. Together they let a [[Multi-Agent System]] span teams and organisations instead of living inside one process.
## Why it matters
Without a shared protocol, every agent-to-agent link is a custom integration. A standard turns a closed [[Orchestrator-Subagent Pattern|orchestrator/subagent]] setup into an open ecosystem where specialised agents can be composed like services.
## Related
- [[Model Context Protocol]]
- [[Multi-Agent System]]
- [[AI Agent]]
- [[Orchestrator-Subagent Pattern]]
## Sources
- A2A (Agent-to-Agent) open protocol specification (2025).