Google A2A Protocol: Connecting Agents
On April 9, 2025, Google announced its new open A2A (agent to agent protocol) with over 50 partners. Google said it may bring AI industry into a new era of agent interoperability. The production-ready version is expected to launch later this year.
What is A2A
A2A(Agent to Agent protocol) is an open protocol that aims to enable AI agents to communicate and cooperate with each other. This fulfills AI agents' capability to better satisfy user demands.
Why A2A
Why do we need AI agents to communicate and cooperate with each other? Understanding might be easier with examples.
You find AI is so smart and capable of many things. Today you want AI to help you check the weather in California for next 7 days, offer some clothing suggestions and add them to your travel plans.
Firstly, there may be 3 AI agents helping you to complete the task:
- Weather agent: Links to the web and searches for weather forecasts
- Clothing advice agent: Provides clothing advice based on the forecast
- Travel agent: Arranges travel plan and manages details of the plan (Why so many agents. Like in real world, let professional handle professional matters 👅)
Then you may follow these steps to complete the task:
- 1.Ask weather agent to search and get the weather information
- 2.Ask clothing agent to provide clothing suggestions based on the weather information
- 3.Ask travel agent to add these information to your travel plan by date
Mission completed 👏.But do you see the problem? You have to engage in multi-turn conversations to gave the instructions to these 3 agents and manage the input to or output from each agent.
These 3 agents can't communicate and cooperate with each other, you have to link the work flow yourself. This will reduce efficiency.
It's time for A2A to show off. The A2A protocol will make it possible for AI agents to communicate and cooperate with each other. Then you don't need 3 steps any more to get weather, clothing suggestions and updated travel plans—you just need to give your final demand to AI and 3 agents would cooperate with each other to update travel plan.
A2A and MCP
MCP(Model Context Protocol) is designed to enable AI model to connect data, tools and resources. It empowers the AI models with the necessary data and tools to complete tasks.
AI bot completes a specific task using data and tools can be considered an AI agent.
“Agents are autonomous applications that can accomplish novel tasks by using tools, reasoning, and user interactions.”
Complex task needs many AI agents. One agent's output is typically another agent's input. The communication and cooperation between AI agents is very important. A2A(Agent to Agent protocol) is designed to focus on this process.
A2A is a complementary to MCP.
Learn more about A2A, check out the official documentation.