A2A 101 ｜ The Universal Language for AI Agents

Introduction

Imagine this scenario: You ask your AI assistant to plan an international trip. It needs to coordinate flight bookings, hotel accommodations, local tour guides, currency exchange, and more. Previously, these services came from different companies using different systems, making it difficult for AI assistants to connect them seamlessly.

The A2A Protocol (Agent-to-Agent Protocol) was created to solve this problem. It's like the "common language" for AI agents, enabling different agents to communicate and collaborate on complex tasks.

1. What is the A2A Protocol

A2A (Agent-to-Agent) is an open standard developed by Google specifically for enabling communication and collaboration between AI agents. Released in 2025, it has gained support from over 150 organizations and is becoming the de facto standard in the AI agent ecosystem.

Understanding A2A in Simple Terms

If MCP (Model Context Protocol) gives AI agents the ability to "use tools," then A2A gives them the ability to "talk to each other."

Think of it this way:

• MCP = USB standard (lets computers connect to peripherals)

• A2A = Internet protocol (lets different computers communicate)

2. Why We Need the A2A Protocol

The Problem

Before A2A, if companies wanted AI agents developed by different teams to work together, they typically needed to:

1. Write extensive custom interface code

2. Resolve compatibility issues between different systems

3. Maintain complex point-to-point connections

This is like each country using its own telephone standard—international calls required specialized conversion equipment.

How A2A Changes Things

3. Core Concepts of A2A

1. Agent Card

Every AI agent needs to publish an "Agent Card"—essentially its digital business card. The card contains:

• Identity: Name, description, provider

• Capabilities: What the agent can do

• Skills: Specific professional abilities (e.g., flight booking, language translation)

• Endpoint: How to connect to the agent

• Security Requirements: What authentication is needed

Discovery: Agent Cards are typically published at the standard path /.well-known/agent-card.json, allowing other agents to automatically discover and understand their capabilities.

2. Task

In A2A, a "Task" is the fundamental unit of work:

• Each task has a unique identifier

• Tasks have states: submitted → working → input-required → completed/failed

• Tasks can involve multi-turn conversations

• Tasks can run for extended periods

3. Message

Agents communicate through messages that include:

• Role distinction: Who sent it (user/agent)

• Content types: Text, files, structured data

• Context: References to previous tasks for continuity

4. Three Interaction Patterns

4. A2A and MCP: How They Work Together

Many people ask: What's the difference between A2A and MCP? Are they competitors?

Answer: They are complementary, working together to build a complete AI agent ecosystem.

Comparison

Real Example: Customer Service System

User → Front Desk Agent (A2A) → Technical Expert Agent (A2A)
                                    ↓
                          Diagnostic Tools (MCP)

• Front desk and technical experts collaborate via A2A

• Technical expert uses MCP to call diagnostic tools

5. Real-World Applications

Scenario 1: Smart Travel Planning

User: "Plan a 5-day trip to Tokyo"

Main Agent → Flight Booking Agent → Hotel Agent → Tour Guide Agent → Currency Agent
         (A2A collaboration for one-stop travel planning)

Scenario 2: Enterprise Smart Office

• Calendar Agent collaborates with Meeting Room Agent

• Email Agent syncs with Task Management Agent

• Expense Approval Agent connects with Finance System Agent

Scenario 3: Cross-Platform Smart Home

Different brands of smart devices work together through unified A2A:

• AC Agent coordinates with Window Sensor Agent

• Lighting Agent works with Security Agent

6. Technical Features of A2A

1. Built on Mature Standards

• Uses JSON-RPC 2.0 as the communication foundation

• Based on HTTP/HTTPS transport protocol

• Supports gRPC and multiple protocol bindings

2. Enterprise-Grade Features

• Authentication: Supports OAuth 2.0, Bearer Token, and more

• Authorization: Role-based access control

• Traceability: Complete task state management and logging

3. Async-First Design

• Native support for long-running tasks

• Human-in-the-loop collaboration support

• Push notification mechanism for real-time updates

7. Quick Start with A2A

Core Steps

1. Publish Agent Card (define capabilities)
2. Implement A2A server endpoints
3. Discover and connect to other agents
4. Send tasks and handle responses

Agent Card Example

{
  "name": "Travel Planning Assistant",
  "description": "Professional travel planning and booking service",
  "capabilities": {
    "streaming": true,
    "pushNotifications": true
  },
  "skills": [
    {
      "id": "flight-booking",
      "name": "Flight Booking",
      "description": "Search and book international flights"
    }
  ]
}

8. Future of A2A

As AI agent technology continues to evolve, the importance of A2A will grow:

1. Ecosystem Expansion: More agent frameworks will natively support A2A

2. Standardization: A2A is poised to become the industry-recognized communication standard

3. Deeper Applications: From single-task collaboration to complex multi-agent systems

Summary

A2A Protocol is critical infrastructure for the AI agent era. It solves the problems of "what to say" and "how to say it" between agents. Through standardized communication mechanisms, AI agents from different sources and architectures can work together like team members, collaborating on complex tasks.

For developers, learning the A2A Protocol lays a solid foundation for building next-generation AI applications. For businesses, embracing A2A means gaining the ability to build agent ecosystems and gaining a competitive edge in the AI space.

Related Resources

• A2A Protocol Official Specification

• MCP vs A2A Deep Dive

• Official Sample Code Repository