What A2A is and why it exists
Agentic systems rarely work in isolation. A complex task — “compare this product across three stores and order the best option” — needs several agents working together: one searches, one compares, one handles payment. Without a common standard, every pair of agents needs its own bespoke integration.
A2A (Agent2Agent Protocol) solves that: Google published an open specification describing how agents discover one another, hand over tasks and return results over a standard interface — regardless of which model or platform sits underneath each of them.
How A2A works
Each agent publishes an agent card — a JSON document listing its capabilities, the operations it supports and its endpoint. An orchestrator reads those cards and decides which agent should receive which subtask.
Orchestrator
├── Agent Card: search-agent → product search
├── Agent Card: compare-agent → specification comparison
└── Agent Card: checkout-agent → order and payment (AP2)
The exchange runs over a standardised API (REST/SSE): the orchestrator sends a task, receives interim updates and finally the result.
A2A vs MCP: two sides of one stack
| Aspect | MCP | A2A |
|---|---|---|
| What it connects | Agent ↔ tools and data | Agent ↔ agent |
| Examples | Search, product database, store API | A search agent, a payment agent |
| Who built it | Anthropic (open) | Google (open) |
| Payment layer | None | AP2 (an extension) |
In practice the two protocols work together: MCP gives an agent its tools, A2A lets agents share complex work between them.
What it means for commerce
A2A opens the way to specialised commerce agents that can operate across the ecosystems of different vendors. A shopper’s own agent will be able to delegate a subtask to a specific retailer’s agent and get a structured answer about assortment, availability and terms — with no custom integration work.
Important: A2A is a standard for the developers of agentic platforms. Through 2025–2026, end retailers are better served by focusing on ACP and YCP — the order-intake protocols — and on the agent-readiness of their catalogue.