Agent-to-Agent Protocol

Agent-to-Agent Protocol
Developer	Google
Status	Stable
First published	April 2025
Transport	HTTPS
Encoding	JSON-RPC 2.0, SSE
License	Apache 2.0
Website	https://google.github.io/A2A
Repository	https://github.com/google/A2A

The Agent-to-Agent Protocol (A2A) is an open standard developed by Google for enabling communication between autonomous AI agents. Launched in April 2025 at Google Cloud Next '25, A2A provides a standardized way for agents to discover each other's capabilities, exchange information, and coordinate actions across platforms.

Overview

A2A addresses the challenge of agent interoperability in enterprise environments where multiple specialized agents from different vendors need to collaborate. It enables agents to work together regardless of their underlying framework or technology stack, building on concepts from traditional Agent Communications Languages.

Design principles

Google designed A2A around five key principles:

1. Embrace agentic capabilities – Enable natural multi-agent collaboration without limiting agents to simple tools
2. Build on existing standards – Use HTTP, SSE, and JSON-RPC for easy integration
3. Secure by default – Enterprise-grade authentication and authorization
4. Support for long-running tasks – Handle tasks that may take hours or days
5. Modality agnostic – Support text, audio, and video streaming

Architecture

A2A follows a client-server model:

Client Agent

The client agent initiates interactions by sending task requests to remote agents. It is responsible for discovering suitable agents, managing authentication, and coordinating workflows.

Remote Agent

Remote agents receive and process task requests. They advertise their capabilities via Agent Cards and handle authorization for incoming requests.

Technical specification

Agent Cards

Agents advertise their capabilities using Agent Cards in JSON format:

Example Agent Card

{{{code}}}

Communication flow

1. Discovery – Client agent identifies suitable remote agents via Agent Cards
2. Authentication – Remote agent verifies client credentials
3. Task execution – Agents exchange messages to complete the task
4. Result delivery – Remote agent returns results to client

Security

A2A supports multiple authentication schemes:

• OAuth 2.0
• API keys
• Mutual TLS (mTLS)
• Scoped tokens for capability-specific access

Adoption

A2A launched with support from over 50 technology partners, including:

• Technology companies: Atlassian, Box, Salesforce, SAP, ServiceNow
• Consulting firms: Accenture, Deloitte, McKinsey, PwC

The protocol is now managed under the Linux Foundation.

Comparison with other protocols

Protocol	Focus	Relationship
MCP	Tool integration	Complementary (vertical vs. horizontal)
ACP	Local-first coordination	Alternative for edge/local scenarios
ANP	Decentralized internet-scale	Alternative with different architecture

See also

• Model Context Protocol
• Agent Communication Protocol
• Agent Network Protocol
• Multi-Agent Systems

External links

• Official specification
• GitHub repository
• Launch announcement