Google to Entrust the A2A Agent Protocol to the Linux Foundation

The future of A2A (Agent2Agent) lies with the Linux Foundation.

Originally conceived as an open protocol for communication between AI agents, Google will transfer its spec, the SDK, and developer tooling to a project with “neutral governance.” It is a founding member alongside AWS, Cisco, Microsoft, Salesforce, SAP and ServiceNow.

A2A positions itself as a complement to MCP (Model Context Protocol), intended to foster the use of tools and data by agents. It involves a central agent (the “client”) that communicates tasks to remote agents (the “servers”). The latter expose their identity, capabilities, and authentication schemes through JSON “visiting cards”.

These “visiting cards” can be discovered in several ways:

• For publicly accessible agents or those broadcast widely, hosted in a well-known folder (RFC 8615)
• In enterprise environments or in specialized ecosystems, use of a registry or catalog
• In development/testing scenarios or for private agents, direct URL-based configuration

Once agents have discovered their respective capabilities, they can negotiate the terms of interaction and coordinate either synchronously (request‑response) or asynchronously (server-side event streaming or push notifications). All of this without exposing their internal state or potentially proprietary tools.

A2A transmits messages in JSON-RPC over HTTP(S). It leverages standard web authentication mechanisms. Microsoft has implemented it in its Azure AI Foundry offering, at the Semantic Kernel layer (its own “LangChain”). Salesforce has recently formalized its integration into MuleSoft, for managing authentication and authorization, the context (notably to prevent prompt injections) and the “visiting cards.” Cisco intends to bridge it with its AGNTCY initiative, launched with LangChain and Galileo (model evaluation) to standardize infrastructure building blocks for the use of AI agents.