M365Con 2027 – כנס אונליין חינמי · 11–22 Jan 2027 · 120 sessions m365con.net

Agent2Agent (A2A) protocol

Concept

A2A (Agent2Agent) הוא פרוטוקול פתוח, המתארח על ידי Linux Foundation, המאפשר לסוכני AI הבנויים על פלטפורמות שונות לגלות זה את זה ולהאציל משימות, באמצעות טכנולוגיות אינטרנט סטנדרטיות כגון HTTP ו-JSON-RPC. הוא משלים את MCP, שמחבר סוכן לכלים ולנתונים.

✦ CONCEPTAgent2Agent (A2A)protocolM365.FM
Title graphic: Agent2Agent (A2A) protocol (Concept)

נושאים שנדונים לעיתים קרובות יחד

בפרקים על Agent2Agent (A2A) protocol עולים לעיתים קרובות גם: Work IQ, Microsoft Entra Agent ID ו-Model Context Protocol (MCP).

Description

The Agent2Agent (A2A) protocol is an open protocol for communication and interoperability between agentic applications that are opaque to one another. It allows AI agents built on different platforms to find each other, delegate sub-tasks, exchange information and coordinate actions. Frameworks named on the project site include LangGraph, CrewAI and Semantic Kernel. The aim is that such agents can connect without a custom integration being built for every pair of systems.

A central design idea is opacity. An agent that takes part in A2A collaboration does not have to expose its internal memory, its tools or its proprietary logic to the agents it works with. Each agent stays a black box that is reached through the protocol, which supports collaboration across vendor and platform boundaries while the implementation details remain private. The project lists secure and opaque collaboration as one of its highlighted features, alongside interoperability, complex multi-agent workflows, working with MCP, and extensibility.

A2A is hosted by the Linux Foundation under neutral governance and is released under the Apache 2.0 license. The project cites more than 170 community partners and says it has joined the Agentic AI Foundation. Changes to the protocol are handled through a formal process. Extensions and custom bindings are added through a tiered promotion process, so the core of the specification can stay stable while new capabilities are developed and tested around it.

For developers, official SDKs are listed for Python, Go, Java, JavaScript, C# and .NET, and Rust. All of them implement the A2A v1.0 specification, which gives teams a shared baseline regardless of the language in which an agent is written. The short description of the protocol also notes that it builds on standard web technologies such as HTTP and JSON-RPC. The A2A landing page itself does not go into discovery or transport details, so those aspects should be checked in the specification and SDK documentation.

A2A is positioned as complementary to the Model Context Protocol (MCP), not as a replacement for it. According to the project site, MCP connects an agent to tools, APIs and data, whereas A2A connects an agent to other agents. The site recommends using both: MCP for giving an individual agent access to the resources it needs, and A2A for letting several agents work together on a larger task. In practice this separates two questions that come up when designing multi-agent systems, namely what an agent can use and who an agent can delegate work to.

Typical situations where A2A is relevant include multi-agent workflows in which agents come from different vendors or open-source frameworks, and organizations that want to avoid writing point-to-point integrations between each pair of agents. Decisions around adoption usually involve which SDK matches the existing stack, how the agent pairs with MCP-based tool access, and whether any needed capability belongs in the stable core or in an extension.

No M365.FM episodes are currently linked to this topic, so this page relies on the public description of the protocol only.

Overview compiled with AI from M365.FM episodes and the facts on this page. You can improve it with the edit button below.

TypeConcept
KindProtocol

Glossary

Community wiki

Short info above, community knowledge below: add practical tips, examples or corrections. New contributions are reviewed first; trusted members publish directly.

Experts and companies for this topic

Nobody has listed this topic yet. Add it to your profile.

Maintained by Mirko Peters (Microsoft MVP) · Last updated: