Google’s Agent2Agent open protocol aims to connect disparate agents

Google has taken the covers off a new open protocol — Agent2Agent (A2A) — that aims to connect agents across disparate ecosystems.

At its annual Cloud Next conference, Google said that the A2A protocol will enable enterprises to adopt agents more readily as it bypasses the challenge of agents that are built on different vendor ecosystems not being able to communicate with each other.

“Using A2A, agents can publish their capabilities and negotiate how they will interact with users (via text, forms, or bidirectional audio/video) – all while working securely together,” said Saurabh Tiwary, vice president of Cloud AI at Google.  

[ Related: Google Cloud Next ’25: News and insights ]

The interoperability being offered by the A2A protocol will allow businesses to automate complex workflows that span multiple systems, potentially increasing productivity while reducing integration costs, according to Paul Chada, co-founder of DoozerAI — an agentic digital worker platform.

Google said the protocol is built atop existing, popular standards including HTTP, SSE, and JSON-RPC. This should make it easier to integrate with existing IT stacks businesses already use.

While HTTP serves as the foundation of web communication, SEE and JSON-RPC are foundational protocols for sending updates to a client server, and applications talking to each other remotely using JSON messages respectively.

“Using HTTP and JSON-RPC is practical and should make implementation easier,” said Anil Clifford, founder of IT services and consulting firm Eden Digital.

[ Related: Agentic AI – Ongoing news and insights ]

However, Clifford was uncertain of the protocol’s success in handling edge cases in real-world scenarios, which he believes will determine the protocol’s proficiency.

In its efforts to proliferate the protocol, Google said that it has partnered with over 50 partners, such as SAP, LangChain, MongoDB, Workday, Box, Deloitte, Elastic, Salesforce, ServiceNow, UiPath, UKG, and Weights & Biases among others.

How A2A works?

A2A facilitates communication between a “client” agent and a “remote” agent.

While a client agent is responsible for formulating and communicating tasks, the remote agent is responsible for acting on those tasks to provide the correct information or take the correct action.

The interaction between agents across use cases is dependent on several key capabilities baked within the protocol, such as capability discovery, task management, collaboration, and user experience negotiation.

The capability discovery feature allows the client agent to read an agent or agents’ capabilities published or showcased via their Agent Cards in JSON format.

This allows the client agent to identify the best agent that can perform a task and leverage A2A to communicate with the remote agent, Google said.

The task management capability allows the client and remote agent to talk to each other to complete a task based on end-user input.

This task object is defined by the protocol and has a lifecycle, Google said. It can be completed immediately or, for long-running tasks, each of the agents can communicate to stay in sync with each other on the latest status of completing a task. The output of the tasks is called an artifact.

The collaboration capability allows the client and remote agent to send each other messages to communicate context, replies, artifacts, or user interactions.

The user experience negotiation capability, according to the executives, allows the client and remote agents to negotiate the correct format needed to respond to an end user request as well as understand the user’s UI capabilities, such as iframes, video, web forms, etc.

Complementary to Model Context Protocol

The A2A protocol is different from Anthropic’s Model Context Protocol (MCP) but both could complement each other, analysts said.

While MCP focuses on the interaction between an application and a generative AI model, the A2A protocol focuses on the interaction between different AI agents.

“You could view MCP as providing vertical integration (application-to-model) while A2A provides horizontal integration (agent-to-agent),” Chada said, adding that an agent built using MCP could potentially use A2A to communicate with other agents.

Chada also pointed out that A2A is distinct from Nvidia’s AgentIQ, which is more of a development toolkit for building and optimizing agent systems rather than an agent communication protocol.

A2A as a new industry standard

Microsoft, Amazon, and other major cloud providers could end up adopting A2A given that major vendors, such as Salesforce, ServiceNow, and Workday are already part of A2A’s partner network, Chada said.

But Clifford feels that Microsoft could have a good reason to develop a competing or complementary standard, given their deep integration with OpenAI and enterprise software presence. “The question is whether enterprises will benefit from this competition or suffer from fragmentation. History suggests we’ll see competing standards before consolidation occurs — likely causing headaches for early adopters,” Clifford said.