The Agent2Agent (A2A) protocol is an open standard that lets AI agents from different vendors and frameworks find each other, share tasks, and exchange results. Google announced it in April 2025, and it is now an open-source project hosted under the Linux Foundation.
It is also written "agent-to-agent protocol." The specification describes its goal as communication "between independent, potentially opaque AI agent systems." Agents work together without needing access to each other's internal state, memory, or tools.
Where the A2A protocol came from
Google announced A2A on April 9, 2025, with more than 50 technology partners. It built the protocol on existing web standards: HTTP, Server-Sent Events (SSE), and JSON-RPC.
On June 23, 2025, the Linux Foundation launched the A2A project. Its announcement calls A2A "an open protocol created by Google for secure agent-to-agent communication and collaboration."
In August 2026, A2A joined the Agentic AI Foundation (AAIF) as a hosted project. The AAIF is itself hosted by the Linux Foundation. The A2A project's announcement names the Model Context Protocol (MCP) among its sibling projects there. As of November 2026, the latest released version of the specification is 1.0.0.
How A2A works
Agent Card. Each A2A server publishes a JSON document that describes the agent: what it can do, where to reach it, and which authentication it accepts. The specification's examples fetch it from a well-known address, /.well-known/agent-card.json.
Tasks. Work is organized as tasks with a lifecycle. A task can be submitted, working, waiting for input or authorization, completed, failed, canceled, or rejected. The protocol supports long-running tasks, including ones that wait for a person.
Messages and artifacts. Agents exchange messages made of parts, such as text, files, or structured data. The output of a task is called an artifact.
Transport. Version 1.0 defines bindings for JSON-RPC, gRPC, and HTTP/REST. Streaming and push notifications carry updates on long-running work.
Security. The specification requires encrypted communication in production (HTTPS, or TLS for gRPC). Servers must reject requests with missing or invalid credentials. Supported schemes include API keys, OAuth 2.0, OpenID Connect, and mutual TLS.
A2A and MCP
The two protocols are often mentioned together. MCP connects an agent to tools and data sources. A2A connects agents to other agents. The A2A project calls MCP the "vertical integration layer" and A2A "the horizontal protocol enabling peer-to-peer collaboration."
A single system can use both. An agent might reach its own systems through MCP and delegate part of a task to another vendor's agent over A2A.
What A2A means for a contact center
A contact center may run a customer service agent from one vendor and a fraud-check agent from another. Its own engineers may have built a third agent for one process. A shared protocol lets them pass work along without a custom integration for each pair. This is the pattern behind BYOA (bring your own agent) and Agent Connect.
The customer still sees one conversation. A few questions are worth settling early: who owns the conversation, how verified identity passes between agents, and where each agent's actions are logged. It also helps to know how an external agent is monitored and who can switch it off.
Frequently asked questions
Who created the A2A protocol?
Google announced A2A in April 2025 with more than 50 technology partners. In June 2025 it became a Linux Foundation project. Since August 2026 it has been a hosted project of the Agentic AI Foundation, which the Linux Foundation hosts.
What is the difference between A2A and MCP?
MCP standardizes how an agent connects to tools, data, and applications. A2A standardizes how agents discover each other and delegate tasks. They cover different connections, so a system can use both.
Is the A2A protocol secure?
The specification sets a baseline: encrypted transport in production, authentication schemes declared on the Agent Card, and rejection of requests without valid credentials. How secure a deployment is depends on the implementation, including token handling, permissions, and logging (see AI agent security).
Do you need A2A to connect an external agent to a customer service platform?
Not always. Platforms can also connect external agents through their own APIs. A2A gives every party the same format, which can reduce integration work when several vendors are involved.
How we approach this at Zowie
Zowie builds AI agents for customer service on voice, chat, and email, with over 100 million conversations a year across our clients. In each conversation, the language model speaks. A separate, deterministic Decision Engine returns every amount, rate, offer, disclosure, and escalation from your rules, with an audit log. To talk through how external agents fit your setup, you are welcome to book a demo.
Explore: Agent Connect, BYOA, Orchestrator, Agentic AI, Agentic workflow, AI agent security