Google Introduces Agent-to-Agent Protocol (A2A) to Break Down AI Silos

“MCP+A2A” becomes the future standard?

huizhou92

— 阅读时间 4 分钟
分享
Google Introduces Agent-to-Agent Protocol (A2A) to Break Down AI Silos

At Google Cloud Next ’25, Google unveiled something that could forever change the landscape of enterprise AI: the Agent-to-Agent Protocol (A2A)—an open, vendor-neutral standard for intelligent agent communication. It is a universal “friend request” protocol for AI agents. And yes, over 50 major companies have already joined.

Announcing the Agent2Agent Protocol (A2A)
Explore A2A, Google's new open protocol empowering developers to build interoperable AI solutions.

A Standard Language for AI Agents

In the past, agents built on different platforms couldn’t easily talk to each other. Enterprises had to deal with data silos and patchwork integrations. A2A changes that by offering a standardized way for agents to collaborate — regardless of vendor, framework, or underlying platform.

Google designed A2A to support major enterprise tools right out of the box; It feels like the early days of Android, when Google rallied dozens of companies behind a shared mobile OS. With over 50 launch partners, A2A is already gaining momentum and setting the stage for a new AI ecosystem.

What Is A2A?

At its core, A2A is an open protocol that allows intelligent agents to interact naturally with one another—even when they don’t share tools, memory, or context.

Imagine a large e-commerce company using:

  • Atlassian for project management
  • Box for file storage
  • Salesforce for CRM
  • Workday for HR

In the past, agents across these platforms were isolated. Now, with A2A, they can securely share data and automate workflows across systems — no custom integrations needed.

Built on Real-World Standards

A2A follows five key design principles:

1. Natural Agent Collaboration

Agents don’t need to share memory or context. A2A enables loose coupling and natural communication between agents — making actual multi-agent scenarios possible.

2. Familiar Tech Stack

It’s built on web standards** like HTTP, JSON-RPC, and Server-Sent Events (SSE). That means it’s easy to integrate into existing enterprise infrastructure without rewriting everything from scratch.

Example: An order management system already using HTTP and JSON-RPC can plug into A2A and instantly access real-time updates from a logistics agent — no new infrastructure is required.

3. Enterprise-Grade Security

A2A supports robust authentication and authorization, aligning with OpenAPI standards. It works seamlessly alongside systems like OpenAI’s ecosystem, allowing secure and compliant data exchange.

4. Flexibility for Every Task

From quick API calls to multi-day research workflows, A2A handles various interaction types. It supports real-time updates and status notifications, so users are always in the loop.

Example: A research institution using A2A to simulate new drug interactions could receive regular updates from agents over a multi-day simulation — like having an AI assistant that reports progress in real-time.

5. Multimodal Support

Agents aren’t just about text. A2A supports audio, image, and video streams, enabling richer interactions and more capable agent systems.

How A2A Works

A2A connects a client agent (the one initiating a task) with a remote agent (the one doing the work). The process looks like this:

🪪 Agent Cards

Each agent exposes a capabilities card—a JSON object describing its capabilities. This helps the client agent find the best match for the task.

🔄 Task Lifecycle

Agents communicate using “tasks,” each with its lifecycle — from assignment to output (called an “artifact”).

Some tasks are completed instantly, while others take time. A2A ensures both agents stay in sync throughout.

🤝 Agent Collaboration

Agents can message each other directly, sharing context, updates, or even partial results. This makes it easier to collaborate on complex, multi-step tasks.

🧩 Experience Negotiation

Each message includes “parts” — content blocks with defined types (like images, forms, and iframes). This allows agents to negotiate the optimal format for the task, adapting to the user’s device and interface.

Who’s Already on Board?

This isn’t just a Google experiment — A2A has already attracted major players across consulting, enterprise software, cloud, and AI:

  • Consulting Giants: Accenture, BCG, Deloitte, McKinsey, PwC, KPMG
  • Enterprise Tech: SAP, Salesforce, Oracle, ServiceNow, Workday
  • AI & Infra: LangChain, MongoDB, Cohere, HCLTech, Infosys

Why It Matters

In recent months, there has been a flood of AI protocols and platforms—some open, many closed. Google’s A2A feels different. It’s not just about connecting models—it’s about creating a shared language for intelligent systems to coordinate, act, and evolve together.

Google has also launched an Agent Development Kit (ADK), a testing tool called Agent Engine, and a brand-new Agent Marketplace, and this is shaping up to be a full-stack ecosystem for agent-based AI.

Compared to past attempts like Microsoft’s MCP, A2A is aiming much higher — and so far, it’s delivering.

Thank you for being a part of the community

Before you go:

阅读更多