Introduction to Agent2Agent (1) - Overview
I found the following article interesting, so I have summarized it briefly.
1. Agent2Agent (A2A)
Today (April 9, 2025), with the cooperation of over 50 technology partners including Atlassian, Box, Cohere, Intuit, Langchain, MongoDB, PayPal, Salesforce, SAP, ServiceNow, UKG, and Workday, as well as major service providers such as Accenture, BCG, Capgemini, Cognizant, Deloitte, HCLTech, Infosys, KPMG, McKinsey, PwC, TCS, and Wipro, we are releasing a new open protocol called "Agent2Agent" (A2A).
"A2A" is an open protocol that complements the "MCP" (Model Context Protocol), which provides tools and context useful for agents. Google designed "A2A" by leveraging internal expertise in scaling agent systems to address specific challenges when deploying large-scale multi-agent systems to users. "A2A" allows developers to build agents that can connect with other agents built using the protocol, enabling users to flexibly combine agents from various providers.
2. A2A Design Principles
A2A is an open protocol that provides a standard way for agents to collaborate, regardless of the underlying framework or vendor. In designing the protocol in collaboration with our partners, we adhered to the following five key principles.
・Leveraging Agent Capabilities
A2A focuses on enabling agents to collaborate through natural, unstructured modalities, even if they do not share memory, tools, or context. It realizes true multi-agent scenarios without limiting agents to just "tools."
・Built on Existing Standards
Because this protocol is built on existing common standards such as HTTP, SSE, and JSON-RPC, it is easy to integrate with the existing IT stacks that companies use daily.
・Secure by Default
A2A is designed to support enterprise-grade authentication and authorization equivalent to OpenAPI authentication schemes at startup.
・Support for Long-Running Tasks
A2A is designed with flexibility to handle all scenarios, from simple tasks to detailed investigations that might take hours or even days if a human were involved. Throughout this process, A2A provides users with real-time feedback, notifications, and status updates.
・Modality Agnostic
The world of agents is not limited to text alone. Therefore, A2A is designed to support various modalities, including audio and video streaming.
3. How A2A Works
A2A facilitates communication between "client" agents and "remote" agents. The client agent is responsible for formulating and communicating tasks, while the remote agent acts on those tasks to provide the correct information or perform the correct actions. This interaction involves several key features.
・Capability Discovery
Agents can advertise their capabilities using a JSON-formatted "Agent Card," allowing client agents to identify the best agent to perform a task and leverage A2A to communicate with the remote agent.
・Task Management
Communication between client and remote agents is aimed at task completion, with agents operating to process end-user requests. This "task" object is defined by the protocol and has a lifecycle. Tasks can be completed immediately, or in the case of long-running tasks, each agent can communicate to synchronize updates on task completion status. The output of a task is called an "artifact."
・Collaboration
Agents can send messages to each other to convey context, replies, artifacts, or user instructions.
・User Experience Negotiation
Each message contains complete content, such as a generated image, called a "part." Each part is assigned a specified content type, so client and remote agents can negotiate the required format and explicitly include negotiations regarding user UI capabilities (such as iframes, videos, web forms, etc.).
For more details on how the protocol works, please refer to the draft specification.