Agent2Agent (A2A) Protocol Documentation
by Linux Foundation
The spec for making agents from different vendors talk — Agent Cards, Tasks and three transports.
Overview
A2A (Agent2Agent) is an open protocol for letting AI agents built by different vendors, on different frameworks, running on different servers, discover each other and collaborate without exposing their internal state, memory or tools. Google published it in April 2025 and donated it to the Linux Foundation; a Technical Steering Committee with members from AWS, Cisco, Google, IBM Research, Microsoft, Salesforce, SAP and ServiceNow now governs it, and the specification has reached version 1.0.0. The documentation site is organised into Documentation (concepts and features), Extensions, the Specification, Resources (SDKs, tutorials and samples), Community and a blog. The specification defines a small data model — an AgentCard describing an agent's identity, skills and service endpoints; Tasks as the unit of work with a defined lifecycle; Messages composed of Parts carrying text, raw bytes or URLs; Artifacts as agent output; and TaskStatus holding state, message and timestamp — plus three transport bindings those objects travel over: JSON-RPC 2.0, gRPC, and HTTP+JSON/REST, with custom bindings permitted. Core methods cover sending a message, streaming a message, getting, listing, cancelling and subscribing to tasks, managing push-notification configuration for long-running work, and fetching an authenticated extended agent card. Security is delegated to standard web schemes: API keys, HTTP Basic and Bearer, OAuth 2.0 with authorization-code, client-credentials and device-code flows, OpenID Connect, and mutual TLS. Official SDKs exist for Python, JavaScript, Java, C#/.NET, Go and Rust; the reference repository is Apache-2.0 with over 25,000 stars.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- ~2-4 hours for the concept docs and tutorials; the specification itself is a 1-2 hour read
- Provider
- Linux Foundation
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Publish an Agent Card advertising an agent's skills, endpoints and security schemes
- ✓Model long-running work as A2A Tasks with a defined status lifecycle and artifacts
- ✓Choose between the JSON-RPC 2.0, gRPC and HTTP+JSON transport bindings for a deployment
- ✓Stream incremental results and subscribe a client to ongoing task updates
- ✓Configure push notifications so a disconnected client learns when a task completes
- ✓Secure agent endpoints with OAuth 2.0, OpenID Connect, mutual TLS or API-key schemes
- ✓Understand where A2A stops and MCP starts when wiring tools versus peer agents
Highlights
- •Vendor-neutral governance: donated by Google to the Linux Foundation with an eight-company steering committee
- •Six official SDKs — Python, JavaScript, Java, .NET, Go and Rust — plus runnable samples
- •Spec has reached 1.0.0, so the data model and method names are stable enough to build against
- •Deliberately opaque-agent design: peers exchange tasks without revealing prompts, memory or tools
- •Read the launch Hacker News thread (450 points, 280 comments) first — practitioners argue hard about whether this duplicates MCP or plain REST, and that scepticism is worth weighing before you adopt
Who It’s For
Best For
- ✓Engineers building multi-agent systems that cross organisational or vendor boundaries
- ✓Platform architects evaluating agent interoperability standards alongside MCP
- ✓Teams exposing an internal agent as a callable service to external partners
Prerequisites
- •Solid grasp of HTTP APIs, JSON-RPC or gRPC, and asynchronous request patterns
- •Familiarity with OAuth 2.0, OpenID Connect and TLS for the security sections
- •Prior exposure to LLM agents and tool calling; MCP knowledge helps but is not required
FAQ
What is Agent2Agent (A2A) Protocol Documentation?
The official documentation and specification for A2A, the open agent-interoperability protocol Google donated to the Linux Foundation. It explains how agents advertise capabilities through an Agent Card, exchange Messages and Artifacts inside long-running Tasks, and stream results or push notifications back — over JSON-RPC, gRPC or plain HTTP+JSON, with standard OAuth, OpenID Connect or mutual-TLS authentication.
Is Agent2Agent (A2A) Protocol Documentation free?
Agent2Agent (A2A) Protocol Documentation is free to access.
What level is Agent2Agent (A2A) Protocol Documentation for?
Agent2Agent (A2A) Protocol Documentation is aimed at a intermediate audience. Recommended background: Solid grasp of HTTP APIs, JSON-RPC or gRPC, and asynchronous request patterns, Familiarity with OAuth 2.0, OpenID Connect and TLS for the security sections, Prior exposure to LLM agents and tool calling; MCP knowledge helps but is not required.
How long does Agent2Agent (A2A) Protocol Documentation take?
Expect roughly ~2-4 hours for the concept docs and tutorials; the specification itself is a 1-2 hour read. Most learners work through it at their own pace.
What will I learn from Agent2Agent (A2A) Protocol Documentation?
You'll learn: Publish an Agent Card advertising an agent's skills, endpoints and security schemes; Model long-running work as A2A Tasks with a defined status lifecycle and artifacts; Choose between the JSON-RPC 2.0, gRPC and HTTP+JSON transport bindings for a deployment; Stream incremental results and subscribe a client to ongoing task updates; Configure push notifications so a disconnected client learns when a task completes; Secure agent endpoints with OAuth 2.0, OpenID Connect, mutual TLS or API-key schemes; Understand where A2A stops and MCP starts when wiring tools versus peer agents.
Topics
Sources
This page was written from 4 sources, 2 on domains other than a2a-protocol.org.