A2A (Agent-to-Agent)
The protocol that gives agents colleagues. Discovery, communication and delegation across boundaries.
The protocol that gives agents colleagues. Agent Cards at well-known URLs let agents discover each other, assess compatibility and delegate state-tracked tasks, with no prior configuration. Created by Google, donated to the Linux Foundation.
A2A (Agent-to-Agent) protocol standardizes how agents discover, communicate and delegate work to each other across organizational and framework boundaries. It was created by Google and, like MCP, donated to the Linux Foundation later on.
A2A works through something called Agent Cards, which are basically JSON manifests served at a well-known URL that describe what an agent can do. Agents discover each other programmatically, assess compatibility and engage without prior configuration. Tasks also have state tracking, thus enabling long-running delegations with real-time updates.
In the emerging three-layer interoperability stack, A2A is the coordination layer that sits alongside MCP: MCP gives an agent capabilities, A2A gives an agent colleagues. Its residence at the Linux Foundation, with broad enterprise backing, signals that the protocol is not going away any time soon and should be considered seriously for adoption.
| Claim | Source | Status |
|---|---|---|
| A2A works through Agent Cards, JSON manifests served at a well-known URL, letting agents discover each other and delegate without prior configuration. | MCP vs A2A: Complete Guide 2026 | verified 2026-07-02 |