"Google version of MCP" is here, open source A2A, Google wins this time

It’s so confusing. “One day of AI is equivalent to ten years of human experience.” This sentence is really not just a slogan.

Just last night, the AI ​​industry released another big news. At the Google Cloud Next conference, Google open-sourced the first standard agent interaction protocol, the A2A Protocol (Agent2Agent Protocol).

So, what exactly is A2A? And what changes can it bring? Let’s find out.

What is A2A Agreement

The A2A protocol mainly solves the communication and interoperability issues between AI agents built by different frameworks and vendors, bringing revolutionary changes to the AI ​​ecosystem. Some people call it the " Google version of MCP ."

I think of it as building an information superhighway for the AI ​​world, allowing AI agents from different "brands/suppliers" to communicate and work together seamlessly .

The A2A protocol can be simply understood as the "common language" between AI agents. Just as the Internet enables people around the world to connect and communicate, A2A enables different AI systems to collaborate across different platforms and frameworks. No matter which vendor these AI agents come from or which underlying technology they use, as long as they support the A2A protocol, they can communicate with each other like old friends.

Why do AI agents need their own communication protocols?

You may ask, don’t we already have communication protocols such as HTTP and RPC? Why do we need a dedicated A2A protocol? This is because the communication between AI agents has its own particularities:

• AI agents need to exchange complex information in multiple formats, including text, images, audio, etc.
• Need to handle long-term tasks, some AI collaborations may take several days
• The need to ensure security and privacy, especially in an enterprise environment
• Need to be able to handle human interaction, such as requesting additional information

The A2A protocol is designed specifically to meet these needs, providing standardized interfaces and formats to make collaboration between AI agents more efficient and secure.

The core design concept of A2A protocol

Google follows several core principles when designing A2A:

1. Simplicity : Built on proven standard technologies such as HTTP, SSE, and JSON-RPC
2. Enterprise-level security : built-in authentication and authorization mechanisms to ensure communication security
3. Asynchronous priority : support long-running tasks and human-computer interaction
4. Modality-agnostic : supports multiple content types, not just text
5. Opaque execution : respects the independence and privacy of each agent and does not force sharing of internal states

Key Components of the A2A Protocol

The A2A protocol consists of several key components:

• Agent Card : A digital identity and capability description of each AI agent, telling other agents what it can do
• Task : A unit of work that needs to be completed, which can be a simple request or a complex project.
• Artifact : The output generated by the task
• Message : Information exchanged between agents
• Part : The basic unit of a message, which can be text, file or structured data

Detailed explanation of core functions

The A2A protocol defines four core functions for agent interoperability:

1. Capability Discovery : This is the process by which agents find each other and learn about each other's capabilities. Agents use Agent Cards in JSON format to "broadcast" their capabilities. Client agents can use this to discover and select remote agents that are best suited to perform specific tasks. Capability discovery is critical to building dynamic, adaptive multi-agent systems.

2. Task Management : This is the core interaction model of the A2A protocol. All communications revolve around the creation, execution, and completion of tasks. The protocol defines clear lifecycle states for tasks, supporting both immediate tasks and long-term tasks that require continuous status updates. The final result of a task is an Artifact.

3. Collaboration : defines how agents exchange information while performing tasks. Agents can share context, send replies, pass products or user instructions by exchanging messages. This enables dynamic collaboration between agents, such as requesting clarification or additional information from each other when needed. This mechanism supports interactions between so-called "opaque agents", that is, agents that do not need to share their internal state or reasoning logic.

4. User Experience Negotiation: Allows agents to adjust the way they interact based on the user's interface capabilities. The protocol uses Parts with content types to achieve this. The client and remote agent can negotiate the required content format, and explicitly support negotiation of user interface capabilities, such as whether iframes, videos, web forms, etc. are supported. The protocol also mentions support for two-way audio/video streaming interactions.

Analysis of application scenarios and use cases of A2A protocol

The A2A protocol has a wide range of application scenarios: The A2A protocol aims to solve several key problems encountered by enterprises when using AI to automate complex processes by promoting interoperability between intelligent agents:

1. Breaking down system silos : Many enterprises have multiple independent applications and data systems. A2A aims to break down these silos and enable agents to communicate and collaborate across system boundaries.
2. Automating complex workflows : Enterprise operations often involve complex, multi-step processes that span multiple departments or systems. A2A is expected to automate these end-to-end workflows by enabling specialized agents with different functions to work together.
3. Empowering professional agents to collaborate : Different agents may have different professional capabilities (such as data analysis, customer communication, task scheduling, etc.). A2A provides a mechanism that allows these professional agents to be combined to complete more complex tasks together.
4. Improve autonomy and productivity : By reducing human intervention and system switching, agents can complete tasks more autonomously, thereby improving overall operational efficiency and productivity.

Typical use cases

Some concrete use cases are provided in the official documentation and partner announcements to illustrate the potential of A2A:

1. Candidate Recruitment Process : This is a detailed example that has been cited many times. A hiring manager can instruct his personal agent through a unified interface (such as Agentspace) to find suitable software engineer candidates based on the job description, location, and skill requirements. The agent then uses the A2A protocol to interact with other agents dedicated to recruiting resources (which may come from different recruitment platforms or service providers) to collect information about potential candidates. After receiving the suggestion, the user can instruct his agent to schedule an interview. After the interview, another agent can also be called through A2A to conduct a background check. This example vividly shows how A2A can coordinate multiple agents to automate a complex, multi-stage enterprise process.

Comparison between A2A protocol and MCP protocol

To understand the A2A protocol more fully, we need to compare it with existing standards, especially Anthropic's Model Context Protocol (MCP):

Future Outlook of A2A Protocol

The launch of the A2A protocol marks an important milestone in the development of AI . As more and more AI systems support the A2A protocol, we will see a truly interconnected AI ecosystem gradually take shape. In this system, AI agents can be freely combined like Lego blocks to jointly solve various challenges facing humanity.