An open standard that lets AI agents built by different companies find each other, describe what they can do and hand work between them. Without a shared standard, getting one company’s agent to delegate a job to another’s needs a custom integration each time. The protocol gives agents a common way to advertise their skills and to pass tasks, progress updates and results back and forth, while keeping each agent’s inner workings private.
Technical view
Introduced by Google in April 2025 and moved to the Linux Foundation in June 2025. Each agent publishes an Agent Card, a JavaScript Object Notation (‘JSON’) document describing its identity, skills, endpoint and authentication requirements. Work is organised as stateful tasks with a defined lifecycle (input-required, completed, failed and others), exchanged over bindings that have grown since launch to include JSON Remote Procedure Call (‘JSON-RPC’), gRPC and Representational State Transfer (‘REST’), with streaming and push notifications. It complements the Model Context Protocol (‘MCP’): MCP connects an agent to tools and data, while this connects agents to one another. A remote agent’s output is still untrusted input, needing injection defences.
Related
Part of AI Terminology Index, section: Agents and Tools.