← AI Terminology
A2A Protocol - Agent-to-Agent
A2A (Agent-to-Agent) protocols define standard ways for autonomous agents to discover, authenticate, message, and hand off tasks to other agents across products and vendors.
They aim to do for agents what HTTP APIs did for services (industry efforts include Google’s A2A and similar interop proposals).
They aim to do for agents what HTTP APIs did for services (industry efforts include Google’s A2A and similar interop proposals).
Why It Matters in AI
Multi-agent products should not be siloed walled gardens. Shared protocols enable specialisation marketplaces and cross-vendor workflows. Alongside MCP (tools/resources), A2A targets agent collaboration itself.
Key Points
| Aspect | Description |
|---|---|
| Goals | Interop, discovery, task delegation, auth |
| Needs | Identity, schemas for tasks/results, streaming updates |
| Risks | Security, prompt injection across agents, accountability |
| Status | Emerging standards landscape (2025+) |
| Vs MCP | MCP connects models to tools/data; A2A connects agents to agents |
| Related | Multi-agent systems, tool use, orchestration |
Simple Analogy
A shared courier language so any company’s delivery robot can hand a package to any other company’s robot at a depot.
Common Usage Examples
- Cross-vendor agent handoffs in enterprise demos
- Task cards with status streaming between agents
- Auth between agent identities
- Compare MCP tool calls vs A2A task delegation
Summary
In short: A2A protocols standardise how agents talk to other agents — interoperability for multi-agent ecosystems beyond single-vendor stacks.