The Google A2A protocol could redefine AI ecosystems for Enterprises

On April 9, Google introduced the Agent2Agent (A2A) protocol, a groundbreaking open standard designed to enable seamless collaboration between AI agents across diverse ecosystems. This announcement marks a pivotal moment for businesses using AI. That’s because A2A eliminates barriers to interoperability, reduces integration complexity, and enables dynamic multi-agent workflows. For enterprises struggling with siloed systems or costly custom integrations, A2A offers a scalable solution that enhances productivity and innovation.

In this article, we examine the significance of the A2A protocol, its technical components, and practical implementation strategies.

What is A2A protocol?

In the simplest terms, A2A introduces a protocol for agents to communicate with each other. This is needed because, right now every AI agent library (like lang-graph, crew-ai, google-adk, smolagents, autogen etc.) gives results in their own different format. This means to use multiple agents together, we need to have bridges to facilitate information exchange between each other.

This requirement creates a huge challenge because we need to build and maintain these bridges, one for each pair of libraries. That’s O(N^2) bridges to maintain. That's silly! We’re forced to use only one library for all our needs, which is not ideal given how this space is still evolving and there is no one gold standard for building agents.

With the introduction of A2A, we now have a standardized way for agents to communicate with each other. We see this is as a step towards a common standard for agent libraries, along with MCP.

Here is a detailed overview of A2A

A2A has two parts—Agent Cards and Comunication. This section explains these elements, and the role they play in seamless collaboration.

Agent Card (full spec)

• This is a JSON object that describes the agent’s skills, input and output formats, authentication mechanism, and capabilities it supports (like streaming, notification, state history).

• his will be used by the clients to discover the agent and its skills.

Communication

There are 5 main components of the A2A protocol:

• TASK

• ARTIFACT

• MESSAGE

• PART

• NOTIFICATION

Task (full spec)

• The core entity that enables clients and agents to work together on specific outcomes.

• Created by the client.

• Status is maintained by the agent.

• They can include messages and artifacts, and maintain status throughout their lifecycle.

Artifact (full spec)

• Immutable outputs generated by agents as task results.

• A single task can produce multiple artifacts, each potentially containing multiple parts (like HTML code and images for a webpage).

Message (full spec)

• Containers for any non-artifact content like user requests, agent thoughts, status updates, and contextual information.

Part (full spec)

• A fully formed piece of content exchanged between a client and a remote agent as part of a Message or an Artifact.

• Each Part has its own content type and metadata.

Notification (full spec)

• A mechanism for agents to communicate task updates to clients, especially for long-running tasks, even when clients are disconnected.

For transport, A2A uses JSON-RPC 2.0 over HTTP. For streaming (if enabled), it uses SSE protocol.

For an example learner, refer to these sample methods and JSON responses to get a taste of A2A in action.

Understanding the mechanics of how A2A works reveals its transformative potential, as both a protocol and a foundation for creating truly autonomous digital teams that can adapt and evolve alongside business needs.

Here’s how to make your agent A2A compatible (Python)

In this section, we provide practical guidance for developers on integrating the protocol into your workflows, including tips on reusing existing system prompts and descriptions.

Reference implementations link: This is necessary as we are going to use a few abstractions from the common module of this reference implementation. Already 3 libraries' code is ported - LangGraph, Crew AI and Google ADK. So, you can use them as references.

Official documentation is still in progress; so wherever possible, we are adding links to the code and explaining the concepts to facilitate your transition.

Follow these steps:

1. Add SUPPORTED_INPUT_TYPES and SUPPORTED_OUTPUT_TYPES field to your agent class, which we will use later.

2. Implement the AgentManager class extending from common.server.task_manager.InMemoryTaskManager.

   Bare minimum functions to implement:

   • `__init__`: Duh! Doy! have self.agent for sure

   • `on_send_task`: It is an async abstract function. Recommended to leave it as async. Implement the whole `graph.invoke()` (for lang-graph) or `crew.run()` (for crew-ai) and other equivalent functions of your respective agentics library in this function.

   Not necessary but very minimal recommendations:

   • `_validate_request`: To validate request types.

     
     

3. A2AServer object, by passing AgentManager object created before in the task_manager parameter of the constructor.

   • AgentCapabilities: Straight forward. refer

   • AgentSkill: You can reuse your existing system prompts and descriptions here [refer]

   • AgentCard: Again just re-use your existing content here. Only new fields to be added are:

     • version: Version of the agent card. This is a string.

     • tags: List of tags for the agent card. This is a list of strings.

     • url: URL of the agent card. This is a string.

Bottom line: implementing A2A is more than a technical upgrade; it’s an opportunity to future-proof your workflows and position your organization at the forefront of AI-driven innovation.

A2A is different from MCP

The A2A protocol and Anthropic’s Model Context Protocol (MCP) address different layers of AI ecosystems. This section clarifies their distinctions and explains how they complement each other in advancing agentic AI systems.

To be sure, the discussion of one standard format for communication and interoperability sounds familiar to MCP. But A2A is not an alternative to MCP. MCP is the emerging standard for connecting LLMs with data, resources, and tools. A2A is a protocol for agents to communicate with each other.

Here is an example to better understand the difference: consider a car repair shop which employs autonomous workers. Workers use special-purpose tools (such as vehicle jacks, multimeters, and socket wrenches) to diagnose and repair problems. The workers often have to diagnose and repair problems they have not seen before. The repair process can involve extensive conversations with a customer, research, and working with part suppliers.

Now let’s model the shop employees as AI agents:

• MCP is the protocol to connect these agents with their structured tools (e.g. raise platform by 2 meters, turn wrench 4 mm to the right).

• A2A is the protocol that enables end-users or other agents to work with the shop employees (“my car is making a rattling noise”). A2A enables ongoing back-and-forth communication and an evolving plan to achieve results (“send me a picture of the left wheel,” “I notice fluid leaking. How long has that been happening?”).

A2A also helps the auto shop employees work with other agents such as their part suppliers.

Using both A2A and MCP, your business can create a layered AI ecosystem where models and agents work in harmony, unlocking unprecedented levels of efficiency and collaboration. Click here to view example credits.

A frontier AI data foundry platform can help you

Every library will create its own implementation of A2A by implementing their compatible AgentManager and A2AServer abstractions. For AgentSkill and AgentCard, most of the system prompts and description that we already have can be re-used except for versioning and tags, so these can be the low hanging fruit for implementing the A2A protocol for your agents.

As businesses prepare to adopt the A2A protocol to unlock seamless AI agent collaboration, Centific’s frontier AI data foundry platform can play a critical role in maximizing its potential. The platform provides a unified infrastructure for curating high-quality data, optimizing workflows, and fine-tuning AI models, which helps ensure that agents powered by A2A operate with precision and efficiency. With Centific’s platform as a foundation, you can fully embrace the transformative power of A2A and drive innovation across industries.

Learn more about the Centific frontier AI data foundry platform.