# What is A2A (Agent2Agent)?

Canonical page: https://agent-protocols-yxt.pages.dev/a2a
Checked against https://a2a-protocol.org/latest/ on 2026-10-20. Uncertain items are marked unverified.

A2A is an open protocol that lets independent AI agent systems talk to each other. A client agent finds a remote agent, sends it work, follows progress and receives results.

## Core concepts

- **Agent Card**: JSON document describing identity, endpoint, capabilities, skills and authentication. The specification (1.0.0, Agent Discovery) lists the well-known URI `https://{server_domain}/.well-known/agent-card.json`, registries or catalogs, and direct configuration as discovery mechanisms. Validate a card: https://agent-protocols-yxt.pages.dev/a2a-agent-card-validator
- **Task**: stateful unit of work with a unique id and lifecycle. States in spec 1.0.0: submitted, working, input required, auth required, completed, failed, canceled, rejected.
- **Message**: one turn with a role (user or agent), made of parts.
- **Part**: text, raw bytes, a URL, or structured JSON data.
- **Artifact**: a concrete output of a task, made of parts.
- **Opacity**: the remote agent is a black box. Memory, plan and tools stay hidden.
- **Transport**: HTTP(S), with polling, Server-Sent Events streaming, and webhook push notifications.

## A2A and MCP

Complementary. MCP is for an agent's tools and data. A2A is between agents. See https://agent-protocols-yxt.pages.dev/a2a-vs-mcp

## Governance

Linux Foundation project. The A2A site also mentions the Agentic AI Foundation; current governing body is unverified.

Demo (simulated, not a production A2A server): https://agent-protocols-yxt.pages.dev/demo