UCP, ACP, MCP in plain words: a guide to agentic-commerce protocols
Updated 2026-08-03
Three protocols people keep mixing up
Agentic commerce comes with three acronyms — MCP, ACP and UCP. They're easy to confuse because all three describe how an AI agent talks to your store. But each one handles a different layer, and knowing the difference helps before a vendor pitches you a "turnkey rollout."
Good news: all three rest on the same foundation, so you can prepare for them with a single set of actions. No hype here — just what they are, how they differ, and what a shop in Moldova or Romania should actually do today.
MCP — how an agent plugs into your data
MCP (Model Context Protocol) is an open standard from Anthropic. Think of it as a USB port for AI: it defines how a language model connects to external tools and data. The model client (Claude, ChatGPT and others) calls an MCP server, and the server hands it tools like "find a product," "check availability," or "build a cart."
MCP itself isn't about payments or shops specifically — it's the universal socket. Through it, an agent reads your catalog the same way it reads the weather or your calendar.
ACP — how a purchase closes inside the chat
ACP (Agentic Commerce Protocol) comes from OpenAI and Stripe. It's about closing the deal: a user tells ChatGPT "buy this," the agent opens a checkout session, and payment clears through a shared payment token. ACP is what powers "Instant Checkout" — buying inside the chat without ever landing on your site.
For ACP a store needs two things: a product feed (what you sell, at what price, in stock or not) and a checkout endpoint the agent hits to create the order.
UCP — a store profile for agents
UCP (Universal Commerce Protocol) is an open protocol whose official specification lives at ucp.dev. A store publishes a profile at /.well-known/ucp listing which capabilities it supports (such as checkout) and where the endpoints are, so the agent knows how to trade with you instead of guessing your site structure.
How they differ
| Protocol | What it handles | Who's behind it | What a store needs |
|---|---|---|---|
| MCP | Connecting an agent to data and tools | Anthropic | An MCP server over your catalog |
| ACP | Checkout and payment in chat | OpenAI + Stripe | Feed and checkout sessions |
| UCP | Advertising what the store can do, and commerce on top of it | Open spec (ucp.dev) | A profile at /.well-known/ucp plus the endpoints it lists |
Simpler: MCP is the wire, ACP is the till, UCP is the sign that tells the agent where the till is and what's on the shelves.
Why it matters already
Honestly: AI-driven orders are barely a thing yet. In issue #1 of our AI Traffic Index (13 Jul – 1 Aug 2026), the panel logged zero AI-attributed orders, while one AI vendor alone sent over 600,000 requests to a single store. Agentic commerce is still under construction, and the protocols are young.
Two reasons to prepare now anyway:
- The plumbing is laid ahead of demand. The very data ACP and UCP need — price, currency, availability in machine-readable form — already helps assistants recommend you today. 90% of human AI visits in our index came from ChatGPT: if it can't read your price, it recommends a competitor.
- The standards converge. MCP, ACP and UCP all rest on one foundation — structured product data. Build it once and you're in all three.
What Botmetria ships today
Preparing for three protocols doesn't mean writing three integrations. Botmetria doesn't run an MCP server or ACP's in-chat Instant Checkout. What it ships — the WooCommerce plugin and SDKs for PHP and Node.js — are three layers built on the same product data:
- A
/.well-known/ucpmanifest — publishes where your catalog, checkout and policies live, in Botmetria's own simplified format rather than the official ucp.dev profile. Botmetria's agent demo reads it; agents that implement only the official spec won't. - A product feed — the same fields as JSON-LD Product:
price,priceCurrency,availability. One source of truth for both your site and agents. - Checkout sessions — an endpoint where an agent builds a cart and gets a link into your native checkout; the shopper places and pays for the order there, and you receive it in your usual system with the source properly attributed.
Where to start
You don't have to roll out all three protocols tomorrow. A sensible order:
- Clean up your product data — JSON-LD Product with price and availability. That's the foundation for all three.
- Make sure bots can see the site at all: images don't return 403/406,
robots.txtdoesn't block AI bots. - Run the free Botmetria audit — it scores your readiness 0–100 and shows what to fix first.
The protocols will keep changing. Clean structured product data won't.
FAQ
How is MCP different from ACP?
MCP is a universal way to connect an agent to data and tools (Anthropic's standard). ACP, from OpenAI and Stripe, handles checkout and payment right inside the chat. In short: MCP is the wire, ACP is the till.
Do I need all three protocols at once?
No. Start with structured product data (JSON-LD Product): price, currency, availability. It's the shared foundation for MCP, ACP and UCP — build it once and you land in all three.
Are there real orders through agentic commerce yet?
Barely — issue #1 of Botmetria's AI Traffic Index (13 Jul – 1 Aug 2026) logged zero AI-attributed orders, while one AI vendor alone sent over 600,000 requests to a single panel store. But it's worth laying the groundwork now, before competitors do.
What is the /.well-known/ucp manifest?
A profile file at /.well-known/ucp that tells agents what the store supports and where its endpoints are; the official format is specified at ucp.dev. The Botmetria WooCommerce plugin publishes its own simplified manifest at that address automatically (catalog, checkout and feed endpoints, shipping and return terms) — it is not the official ucp.dev profile, so agents that implement only the official spec won't read it.
Related guides
Check your store for free
A 60-second audit shows how ready your store is for AI agents — and exactly what to fix first.