A2A Protocol

Definition

The Agent2Agent (A2A) Protocol is an open communication standard that lets autonomous AI agents built on different frameworks and by different vendors discover each other's capabilities, exchange messages, and collaborate on multi-step tasks.

Overview

As organizations move from single chatbots to networks of specialized AI agents, a new problem emerged: agents built with different frameworks had no common language for talking to one another. A2A was created to solve exactly that — it defines a standard way for an agent to publish an "agent card" describing what it can do, and for other agents to send it tasks, stream back results, and negotiate multi-turn collaboration over plain HTTP and JSON-RPC. A2A is deliberately complementary to tool-calling standards rather than a replacement for them: where a model-to-tool protocol lets a single agent call a function or database, A2A operates one layer up, letting one agent treat another whole agent as a remote collaborator with its own reasoning, memory, and tools. This matters for enterprise workflows where, say, a scheduling agent, a research agent, and a customer-support agent — built by different teams on frameworks like LangChain or AutoGen — need to hand off work to each other without every team agreeing on one internal framework. The protocol grew out of work led by Google alongside a large group of enterprise and software partners, and governance was subsequently opened up under a neutral foundation to encourage broad, vendor-independent adoption. It is still an early, fast-moving standard, and the courses on AI Agents & Agentic Workflows and the post on how AI agents actually work are useful starting points for understanding where protocols like A2A fit into the broader agentic stack.

Specification

• Agent cards — standardized JSON metadata describing an agent's skills, inputs, and endpoints
• Task-based interaction model with support for long-running, asynchronous work
• Streaming updates so a calling agent can monitor progress on multi-step tasks
• Transport over standard HTTP and JSON-RPC, avoiding proprietary connectors
• Framework-agnostic design that works across vendors and internal tooling
• Built-in support for multimodal payloads such as text, files, and structured data
• Open governance intended to keep the spec vendor-neutral as adoption grows

Use Cases

History

The Agent2Agent (A2A) Protocol is an open standard for interoperability between AI agents, letting agents built by different vendors and on different frameworks discover one another, delegate tasks, and coordinate work across systems. It was announced by Google on April 9, 2025, and released under the Apache-2.0 license with more than 50 launch partners including Atlassian, Box, LangChain, MongoDB, Salesforce, SAP, and ServiceNow. A2A deliberately reuses familiar web standards — HTTP, Server-Sent Events, and JSON-RPC 2.0 — and introduces "Agent Cards" by which an agent advertises its capabilities. Governance of the protocol was subsequently placed under the Linux Foundation as a neutral home, positioning A2A as a foundational interoperability layer for multi-agent systems, complementary to the Model Context Protocol.

Frequently Asked Questions

Related in Web