A2A
Agent-to-Agent (A2A) is a communication protocol for multi-agent system collaboration, designed to standardize task delegation, information exchange, and coordinated execution between AI agents. The protocol defines standard interfaces and message formats for inter-agent interaction, enabling heterogeneous agents to discover one another and collaborate across platforms and frameworks. It is typically used alongside tool-calling protocols such as MCP to form the foundational communication layer of multi-agent applications.
Core Facts
A2A (Agent-to-Agent) is an inter-agent communication protocol proposed by Google.
A2A introduces the concept of an Agent Card, where each Agent publishes a structured capability declaration document describing the task types it can handle, input/output formats, and authentication requirements.
The A2A protocol focuses on peer-to-peer collaboration between agents, while Anthropic's MCP focuses on connecting agents to tools and data sources.
Timeline (last 90 days)
Poppy, MCP, and A2A are not substitutes for each other but protocols at different layers: Poppy determines which Agent is making the request and what the customer has allowed, MCP servers are one of the channels for Poppy-routable Agents, and A2A handles communication between independent Agent systems.
The security toolchain surrounding agent protocols such as MCP and A2A will gradually mature, covering dimensions including capability inventory management, privilege minimization, access review, and behavioral monitoring.
In multi-agent systems, a low-privileged entry agent may delegate tasks to higher-privileged backend agents via the A2A protocol, causing implicit privilege propagation—a risk that is invisible in static permission inventories.
A developer built a tool called Agent Protocol Inspector that generates structured IGA-style review documents for agent capabilities exposed under the MCP and A2A protocols.
A2A is a unidirectional client-server model; to achieve bidirectional communication, both the client and server must be implemented on the same side simultaneously.
MCP was proposed by Anthropic, A2A was proposed by Google and is still in an early draft stage, and ACP is led by IBM.
The second agent used the A2A protocol to read Dexto's agent card, encapsulated the purchase action as a task with an x402 payment attached, and paid 10 cents in USDC on the Base chain.
After developers connect agents to Dexto via MCP or A2A, they can access the issuance market or the purchase market, with token issuance and trading corresponding to two different connection protocol paths respectively.
Dexto is a middleware layer that both provides an MCP server for agents to connect and issue tokens, and exposes an agent card via the A2A protocol for other agents to read and trade.
WorkIQ offers three access methods: REST API, A2A (Agent-to-Agent) protocol, and MCP (Model Context Protocol).
32 more timeline events
All Facts (20)
A2A (Agent-to-Agent) is an inter-agent communication protocol proposed by Google.
80%VerifiedA2A introduces the concept of an Agent Card, where each Agent publishes a structured capability declaration document describing the task types it can handle, input/output formats, and authentication requirements.
80%VerifiedThe A2A protocol focuses on peer-to-peer collaboration between agents, while Anthropic's MCP focuses on connecting agents to tools and data sources.
80%VerifiedGoogle released the A2A (Agent-to-Agent) protocol in April 2025, specifically to address collaboration issues in multi-agent systems.
80%VerifiedMCP addresses the connectivity problem (how a model reaches tools), while Skill addresses the usage problem (when and with what strategy the model combines tools).
75%VerifiedProtobuf is a language-neutral, platform-neutral structured data serialization protocol developed by Google, offering smaller data size and faster parsing speed compared to JSON and XML.
65%UnverifiedA2A协议的核心设计原则是Agent之间基于声明的能力来协作，而不是互相检查对方的内部实现
95%UnverifiedAgent Card是A2A协议中智能体的标准化名片，以JSON格式声明能力、输入输出模式、认证方式等，但不暴露内部模型和工具实现
95%UnverifiedA2A协议的核心工作单元是Task，具有唯一ID和完整状态机，包括已提交、工作中、已完成、已失败、已取消和需补充信息等状态
95%Unverified文章介绍的六大AI智能体协议为MCP、A2A、UCP、AP2、A2UI和AGUI
95%UnverifiedA2A协议通过Agent Card让每个Agent暴露能力清单，实现Agent之间零代码、零API契约的即插即用连接
90%UnverifiedGoogle创建了A2A（Agent-to-Agent）协议并将其捐赠给Linux基金会
90%UnverifiedA2A协议的传输层使用JSON-RPC 2.0走HTTPS，长任务用Server-Sent Events做流式更新或Webhook做异步通知，版本0.3还加了gRPC支持
90%UnverifiedA2A协议基于HTTP/SSE构建，引入了Agent Card概念，每个Agent发布描述自身能力的元数据卡片
90%UnverifiedA2A和MCP是互补关系而非竞争关系，底层用MCP连接工具和数据，上层用A2A做多Agent编排
85%UnverifiedA2A和MCP在2025年底被捐赠给Linux Foundation旗下的Agentic AI Foundation共同治理
80%UnverifiedA2A协议在2026年4月正式推至1.0版本
80%UnverifiedMCP解决Agent连接工具的垂直问题，A2A解决Agent连接Agent的水平协作问题，两者互补而非竞争
80%Unverified在ADK中，用to_a2a方法一行代码就能将任何智能体暴露为A2A服务
80%UnverifiedSpring AI Agent Utils支持基于A2A协议的多Agent编排协作能力
70%