ACP, UCP, AP2, MCP — the agentic-commerce protocols, in plain English.
Four acronyms, a lot of confusion, and one reassuring truth: you don't implement any of them. Here's what each protocol is, who's behind it, and the only thing you actually have to do.
The short answer
ACP and UCP are how AI discovers and checks out your products. AP2 is how agents pay. MCP is the underlying agent-to-tools plumbing. The platforms implement all four — you just have to be ready for them.
The four protocols
What each one is — and what it means for you.
Backed by OpenAI (with Stripe)
The open standard behind buying inside ChatGPT — how a merchant's products and checkout are exposed to an AI so a shopper can purchase without leaving the conversation.
For you: Your products become buyable in ChatGPT when your catalog/feed is eligible and complete. You don't hand-code ACP — your platform and feed do the work.
Backed by Shopify + Google (and partners)
The discovery + transaction standard for agentic storefronts — how AI agents find, understand and act on a store's catalog across surfaces like Google's AI shopping and Shopify's agentic storefronts.
For you: On Shopify this is largely automatic (agentic storefronts, agents.md, discovery). Your job is catalog quality and schema so the agent has something good to surface.
Backed by Google (and payment partners)
A standard for how an AI agent authorizes and settles a payment on a shopper's behalf — the 'can this agent actually pay' layer beneath agentic checkout.
For you: Payments plumbing handled by the platforms and processors. You care that your checkout is agent-compatible, not that you implement AP2 yourself.
Backed by Anthropic (widely adopted)
The general standard for connecting AI models to tools and data. Not commerce-specific, but it's how agents plug into systems — increasingly relevant for exposing store data and actions to AI.
For you: Mostly a developer/integration concern today. It matters when you want agents to read structured store data or trigger actions — an area to watch, not act on yet.
Emerging standards — details shift as the platforms iterate. We track the specs so you don't have to.
The only thing you actually have to do
Under every protocol, the lever is the same: a complete, machine-readable, schema-rich catalog; server-rendered facts; agent-compatible checkout; off-site authority; and measurement. Get those right and you're ready for whichever acronym wins.
Questions we get a lot.
What is the difference between ACP, UCP, AP2 and MCP?
ACP (OpenAI/Stripe) makes products buyable inside ChatGPT. UCP (Shopify/Google) is how agents discover and transact with a storefront across surfaces. AP2 (Google) is the payment-authorization layer for agents. MCP (Anthropic) is the general model-to-tools standard, not commerce-specific. In short: ACP and UCP are commerce discovery/checkout, AP2 is payments, MCP is the underlying agent-integration plumbing.
Do I need to implement these protocols myself?
Almost never. The platforms (Shopify, Google, OpenAI, payment processors) implement the protocols. Your job is to be ready for them: a complete, schema-rich, machine-readable catalog and an agent-compatible checkout. Getting the acronyms right matters far less than getting your store recommendable.
Which one matters most for a Shopify or Amazon merchant?
For Shopify, UCP-era agentic storefronts + ACP (ChatGPT checkout) are where the near-term revenue is; for Amazon it's Rufus/COSMO readiness (a separate track). In every case the lever is the same: catalog quality, schema, and cross-surface authority — which is exactly what an agentic-commerce readiness program delivers.
The protocols keep changing — how do I keep up?
You don't have to track every spec revision. Focus on the durable fundamentals (machine-readable catalog, complete schema, server-rendered facts, off-site authority, measurement); those pay off under every protocol. Run the free AI Shelf Check to see where you stand today.
Skip the acronym anxiety. See if AI recommends you today.
The protocols are the platforms' job. Being the recommended product is ours. Run the free AI Shelf Check for your baseline, then let's close the gap.