A2A Protocol: Making AI Agents Talk to Each Other

You have an AI that writes articles, an AI that creates images, and an AI that sends emails. You want them to work together—after the writing AI finishes, it notifies the image AI to create visuals, which then notifies the email AI to send. But how do the AIs communicate with each other? A2A is the answer.

💡 What You Will Learn

You have an AI that writes articles, an AI that creates images, and an AI that sends emails. You want them to work together—after the writing AI finishes, it notifies the image AI to create visuals, w

📜 Table of Contents

• A2AWhat Is
• Core Concepts
• Agent Card
• Task
• MCP vs A2A

A2AWhat Is

Core Concepts

Agent Card

{
  "name": "",
  "description": "Generate",
  "skills": ["image-generation", "image-editing"],
  "url": "http://draw-agent:8080/a2a"
}

Task

{
  "task_id": "task-001",
  "type": "image-generation",
  "input": {
    "prompt": "",
    "style": ""
  }
}

Agent A
  ↓ Agent B
  ↓ Task
Agent B
  ↓ 
  ↓ URL
Agent A
  ↓ Embedding
  ↓ 

# Agent A  Agent B
import requests
import json

# 1. Agent
response = requests.get("http://draw-agent:8080/.well-known/agent.json")
agent_card = response.json()
print(f"Agent: {agent_card['name']}")
print(f": {agent_card['skills']}")

# 2. 
task = {
    "task_id": "img-001",
    "type": "image-generation",
    "input": {"prompt": "AI"}
}
response = requests.post(
    "http://draw-agent:8080/a2a/tasks",
    json=task
)
print(f": {response.json()}")

# 3. 
result = requests.get(f"http://draw-agent:8080/a2a/tasks/img-001/result")
print(f"URL: {result.json()['output']['image_url']}")

MCP vs A2A

|:----|:----|:-----| | A2A | AI → AI ||