Google has introduced a new version of its open source Agent2Agent (A2A) protocol, Version 0.3, with capabilities such as gRPC support, enhanced security, and developer-focused capabilities that the hyperscaler said will make it easier for enterprises to integrate the protocol into their agent-related workflows.

A2A now supports gRPC, another open source framework developed by Google that enables applications to communicate with each other via remote calling methods. gRPC is designed to run in a highly performant manner, even when the applications are on different machines or written in different languages.

gRPC’s high performance and support for multiple programming languages make it suitable for complex, distributed applications, said Stephanie Walter, analyst at HyperFRAME Research.

The framework’s low-latency and high-throughput abilities when integrated inside A2A will make it more adaptable for real-time multi-agent orchestration, according to Dion Hinchcliffe, lead of the CIO practice at The Futurum Group.

“For enterprises, gRPC support will allow agents to interoperate over a widely adopted, language-agnostic protocol. This simplifies integration with microservices and existing cloud-native architectures,” Hinchcliffe said.

In its efforts to boost enterprise adoption, Google has introduced a new security capability — the ability to sign security cards — to A2A.

This ability is critical for developers and large enterprises, especially Fortune 500 companies, as they won’t deploy agents that can’t prove their identity cryptographically, said Paul Chada, co-founder of DoozerAI — an agentic digital worker platform.

Seconding Chada, HyperFRAME’s Walter said that the new security capability will help enterprises ensure that appropriate access control and runtime policies are followed, protecting against an attack that could potentially damage reputation, expose trade secrets, or negatively affect the bottom line.

Developers, too, can use the capability to ensure that any agent, especially ones not written by the enterprise, comes from a trusted source, Walter added.

A2A integration with Agent Development Kit

Google has also integrated A2A with its open source framework for building agents — Agent Development Kit (ADK), and Walter feels that this will accelerate agent integration and composability for enterprises using ADK.

“What Google is actually doing is building A2A protocol support directly into ADK, so agents built with ADK automatically get A2A communication capabilities. Think of it like building Slack integration directly into your development framework rather than requiring separate integration work,” Doozer’s Chada explained.

The hyperscaler has also extended client-side support in the Python SDK integrated with its Agent Development Kit.

This extension will make it easier for developers to create and manage A2A agents using Python, Hinchcliffe said.

“For enterprises, it reduces development friction, allowing teams to build, test, and deploy agentic workflows faster while also staying within their existing AI toolchain,” Hinchcliffe explained.

In order to boost the proliferation of A2A, Google will now allow its partners to sell A2A-supported agents on its AI Agents Marketplace. It is also passing on the ability to evaluate agent systems that support A2A via the Vertex GenAI Evaluation Service.

A2A versus MCP: Which one is right for you?

When it comes to the two open source protocols that allow agents to talk to each other, experts say that the Model Context Protocol (MCP) is ahead of A2A.

“From what we are seeing in the developer community, MCP has momentum on ease of use and has great community adoption,” Doozer’s Chada said.

MCP also has broader vendor support and is more ecosystem-neutral, according to Hinchcliffe.

Google said it currently has at least 150 partners who are helping it build, codify, and adopt A2A as the standard protocol for AI agents.

However, Chada and Hinchcliffe are divided in their opinion of which use cases suit each protocol.

At one end, Hinchcliffe said that A2A is more suited for agent ecosystems tied to Google’s stack, and MCP is more suited for multi-vendor, heterogeneous environments where enterprises need interoperability across different models and vendors.

On the other hand, Chada noted that MCP excels in single-agent tool integration scenarios, while A2A targets multi-agent orchestration that enterprises actually need.

“MCP is winning the developer mindshare race, but A2A has the enterprise partnerships that matter for large-scale deployments. The question is what users want: simplicity or enterprise features,” Chanda said.

But both experts agree that A2A is more secure than MCP. “A2A currently offers stronger built-in security primitives: signed security cards and Google’s zero-trust backbone. This makes it easier for enterprises to enforce secure agent interactions out of the box. MCP’s security is more flexible but requires careful implementation across vendors, which can introduce gaps,” Hinchcliffe said.