Glossary Entry

Agent2Agent Protocol

An open protocol, announced by Google in April 2025, for one AI agent to discover another, hand it a task, and track that task to completion without either sharing its internal tools or memory.

LLMs Deployment

Also called: A2A, Agent2Agent

Seed source: A2A specification

Each A2A agent publishes an Agent Card describing its capabilities, and work moves through a task with defined states such as working, input required, completed, and failed. Agents are deliberately opaque to each other, which suits collaboration across company boundaries. Version 1.0, the first stable release, shipped in March 2026.

The project describes the split with the Model Context Protocol as “MCP inside agents, A2A between agents.”