Agent-to-Agent Protocol (A2A)

What is the Agent-to-Agent Protocol (A2A)?

The Agent-to-Agent Protocol (A2A) is an open standard that allows AI agents built by different companies or using different technologies to communicate securely and work together. Originally created by Google and now managed by the Linux Foundation [2][3], A2A provides the connective tissue for multi-agent systems where specialized agents need to coordinate.

The Analogy: A Universal Language for Specialists

Imagine a construction site with an architect, a structural engineer, and an electrician. Each is an expert, but they trained at different schools and use different jargon. If they cannot communicate, the building never gets finished.

A2A is their shared language. It does not change how the electrician does their wiring or how the architect draws blueprints. Instead, it provides a standard way for them to exchange information, request help, and coordinate without needing to understand each other’s internal methods.

How A2A Works: Agent Cards and Discovery

At the heart of A2A is the Agent Card, a JSON metadata document that an agent publishes to describe its identity, capabilities, skills, and how to reach it [1]. Think of it as a digital business card. Agent Cards enable capability-based discovery, meaning agents can find each other and decide whether to collaborate based on declared skills, without exposing internal state, private memory, or proprietary tools.

When agents connect, they use JSON-RPC 2.0, a lightweight standard for one program to call functions on another, carried over the web (HTTP/HTTPS) [1]. For real-time exchanges, A2A supports Server-Sent Events (SSE), a technique that lets one side push updates to the other as they happen, enabling smooth streaming conversations [1]. Agents can interact in three modes: synchronous request-response, streaming, or asynchronous push notifications for long-running tasks.

A2A vs. MCP: Horizontal vs. Vertical

A2A is often discussed alongside the Model Context Protocol (MCP). They serve complementary roles:

• MCP (Vertical): Connects an agent to its tools and data sources, like databases, APIs, or file systems. It is the bridge between the agent and the resources it depends on.
• A2A (Horizontal): Connects agents to other agents. It is the bridge between different AI entities that need to collaborate on complex, multi-step work.

They are complementary, not competing. An agent might use MCP to pull data from a database and then use A2A to send that data to a specialized analysis agent. Complex agentic AI systems typically use both.

Why A2A Matters for Agents

As AI systems evolve from single models into multi-agent systems, the ability for agents to discover and collaborate becomes foundational. A2A enables a pattern where specialized agents, each with narrow expertise, can split work and share results to accomplish goals too complex for any single agent.

Backed by over 150 organizations including AWS, Google, IBM, Microsoft, Salesforce, SAP, and ServiceNow [2], A2A is integrated into major cloud platforms. Because it is an open standard under the Linux Foundation [2], it ensures that multi-agent collaboration is not locked into a single vendor’s ecosystem. For orchestration frameworks managing multiple agents, A2A provides the interoperability layer that lets heterogeneous agents work together.

Limitations

A2A is a communication protocol, not an intelligence layer. It standardizes how agents talk to each other but does not solve challenges like trust between agents, quality control over delegated work, or security in adversarial multi-agent environments. As with any open standard, adoption varies across vendors and frameworks, and the ecosystem is still maturing. The protocol does not define how agents should be governed or how conflicts between agents should be resolved.

Reader Questions

Does A2A expose my private data?

No. A2A is designed so agents can discover and collaborate without exposing internal state, memory, or proprietary tools. Agent Cards share only what is needed for the interaction [1].

Is A2A a competitor to MCP?

No. A2A and MCP serve different layers. MCP connects agents to tools (vertical). A2A connects agents to other agents (horizontal). Most agentic systems use both.

Who maintains A2A?

A2A is an open standard managed by the Linux Foundation [2]. It was originally created by Google [3] and is supported by a broad coalition of over 150 organizations [2].

Can agents from different vendors use A2A?

Yes. That is the core purpose. A2A enables agents built on different frameworks and by different companies to interoperate, as long as they implement the A2A specification [1].