xDev Asia

Google A2A Protocol: Agent Cards, capability discovery, task lifecycle, inter-agent communication. Compare A2A vs MCP. Demo two agents from two different frameworks collaborating.

Build AI Agents: From Zero to Production

Part 5: MCP, A2A & Multi-Agent Systems

xdev.asia

Introduction

MCP connects the agent to tools/data. But what about when you need the agent to connect to another agent? A2A (Agent-to-Agent Protocol) — developed by Google — solves this problem.

1. A2A vs MCP

	MCP	A2A
Connect	Agent ↔ Tools/Data	Agent ↔ Agent
Model	Client/Server	Peer-to-Peer
Discovery	Tool schemas	Agent Cards
Use cases	Read DB, call API	Delegate tasks between agents

2. Agent Cards

{
  "name": "Research Agent",
  "description": "Tìm kiếm và tổng hợp thông tin",
  "capabilities": ["web_search", "summarize", "translate"],
  "endpoint": "https://research-agent.example.com/a2a",
  "input_schema": {...},
  "output_schema": {...}
}

3. Cross-Framework Communication

A2A allows agents from LangGraph to communicate with agents from CrewAI, AutoGen, or any other framework.

Summary

A2A = protocol for agent-to-agent communication
Agent Cards = self-description for capability discovery
MCP + A2A = full connectivity (tools + agents)
Cross-framework interoperability

Exercises