A2A

live

Agent2Agent (A2A) Protocol

by Google· launched 2025-04verified 2026-09-02

At a glance

Layer
communication
Category
framework
Rails
—
Settlement
—
Reversible
—
Human in loop
—
D2C fit
—
B2B fit
—
Spec
https://a2a-protocol.org/latest/

An open interoperability layer for agent-to-agent communication: it lets independent agents discover each other, exchange capabilities, and coordinate tasks. It is not a payment-execution protocol itself — payments ride on top via other protocols.

How it works

A2A is a communication and interoperability standard, not a payment rail. Its purpose is to let agents built by different vendors, on different frameworks, discover one another and collaborate on tasks.

  • Agent Cards. Each agent publishes a machine-readable descriptor advertising its identity, skills, and endpoints, so a client agent can find and evaluate a remote agent.
  • Tasks & messages. A client agent opens a task with a remote agent and they exchange structured messages and artifacts until the task completes, supporting long-running and streaming interactions.
  • Transport. It layers on ordinary web primitives (HTTP, JSON-RPC/SSE), which keeps it framework-agnostic and easy to adopt alongside existing services.

The most common point of confusion is treating A2A as a way to pay agents. It is not: A2A carries the coordination, and value transfer is delegated to a dedicated payment or authorization layer (for example AP2 for mandates, or an execution protocol such as x402). Keeping this distinction explicit is essential when reading the comparison matrix — the settlement and reversibility columns are marked not-applicable for A2A on purpose.

Tradeoffs

  • Carries no payment semantics: any monetary exchange must be delegated to a separate authorization/execution protocol.
  • Interoperability depends on both agents adopting the same version and honoring published Agent Card capabilities.
  • As a coordination layer, it inherits the trust and identity problems of whichever agents connect through it.

Security & limitations

  • Agent discovery and capability advertisement widen the surface for impersonation and capability spoofing without strong agent identity.
  • Because tasks can be long-running and delegated, injection or manipulation mid-task can propagate to downstream systems the agent can reach.

See sources below. Documented for commentary and identification only.

Fits together with

  • AP2 — Agent Payments Protocol (AP2)
  • x402 — x402 Payment Protocol

Deployments using A2A

No deployments recorded yet.

Sources

  1. [1]A2A Protocol — documentation— A2A Project
  2. [2]A2A: A new era of agent interoperability— Google Developers Blog, 2025-04
  3. [3]A2A specification and reference material— A2A Project (GitHub)