ACP, UCP, AP2 and MCP Explained for Merchants
A plain-language guide to the agentic-commerce protocols merchants keep hearing about, what each one does, why it matters for your catalog, and what to prepare now without betting on a winner.
If you sell online, four acronyms keep coming up — ACP, UCP, AP2 and MCP — and the natural question is which one you should build for. The honest answer: do not bet your business on a single protocol winning. Make your product truth certified and machine-readable, keep control of your own checkout and customer data, and you will be ready for whichever standards end up mattering.
Here is what each one is, the problem it solves, and why it touches your catalog and checkout. Treat the specifics as a moving target — these standards are young and evolving — but the shape of the problem each addresses is stable enough to plan around.
The four protocols, in plain terms
ACP — the assistant-to-merchant feed
ACP (Agentic Commerce Protocol), associated with OpenAI and Stripe, is a way for an AI assistant to read a merchant's products and, in some flows, pass an order and payment through to the merchant. Think of it as the interface between a large consumer assistant and your store: the assistant needs a clean, structured description of what you sell so it can present the right item to a shopper who is researching inside the chat.
The problem it solves is presentation and hand-off — how an assistant surfaces your product and routes a ready-to-buy shopper without scraping your site and guessing. For a merchant, the practical implication sits upstream of the protocol: the assistant can only be as accurate as the product data it is handed. Complete, correct attributes and claims are what make you eligible to appear well; thin or scraped data is what gets you misrepresented or skipped.
UCP — the shared checkout and discovery vocabulary
UCP (Universal Commerce Protocol) aims to be a common language for the whole commerce lifecycle: how products are discovered, how a cart is built, how checkout and orders are handled across different agents and platforms. Where ACP is tied to one assistant ecosystem, UCP is pitched as neutral plumbing that many players can adopt.
The problem it solves is fragmentation. Without a shared vocabulary, every assistant and marketplace would need a bespoke integration with every store. UCP standardizes the handshake. For your catalog, the takeaway is the same as ACP's: the protocol defines the envelope, not the contents. It standardizes how an agent asks "what do you sell and can I buy it," but not whether your answer is complete, current and true. And critically, a shared checkout vocabulary does not mean you surrender the checkout — you can speak the standard while keeping yourself as merchant of record.
AP2 — proof that the shopper actually authorized the purchase
AP2 (Agent Payments Protocol), associated with Google, tackles a narrower and deeper question: when an agent pays on a shopper's behalf, how does everyone prove the shopper truly authorized that specific purchase? It uses signed "mandates" — cryptographic evidence of the shopper's intent and of the exact cart — so a payment can be trusted and disputed cleanly.
The problem it solves is trust and liability in delegated payments, which is the piece that has to be neutral for agent-driven buying to be safe at scale. Most merchants do not implement this by hand; you consume it through your payment and checkout stack. Why it matters to you is indirect but real: it lets banks, networks and platforms accept an agent's purchase as legitimate — a precondition for any of the shopping flows above to turn into money.
MCP (Model Context Protocol), from Anthropic, is the most foundational and the least commerce-specific. It is an open standard for how an AI model connects to external tools and data sources — a universal port that lets an assistant call a function or read a structured resource, whatever the vendor. Commerce protocols increasingly ride on top of it.
The problem it solves is connection. Before a shared standard, every model-to-tool link was a one-off. MCP makes your data and capabilities reachable by assistants in a consistent way. For a merchant, MCP is the reason a certified, machine-readable catalog is so valuable: if your product truth is exposed as a clean, queryable resource, assistants can consume it through the same standard everything else is converging on — rather than guessing from a rendered web page.
Why "don't pick a winner" is the right posture
Notice the pattern across all four: every protocol standardizes the pipes, and none of them fills the pipes with good data on your behalf. ACP and UCP define how an agent asks about and buys your products, AP2 defines how the payment is trusted, and MCP defines how the connection is made. What none of them do is decide whether the description of your products is accurate, up to date, and safe to repeat.
That is the part you own, and it is also the part that carries across whichever standard wins the next round. A market example makes the risk concrete: some large retailers wall their assistants off entirely — Amazon's shopping assistant, renamed from Rufus to Alexa for Shopping in May 2026, is only reachable inside Amazon — so no single external protocol reaches every surface anyway. Building for one and hoping is a poor hedge.
What to do now
Get your product truth into a form that any of these standards can use. In Querytail terms, that means three things:
- Certified Catalog. Turn your existing catalog into certified, machine-readable Agent Cards — structured product truth your team reviews and approves, so what an assistant reads about your products reflects what you actually stand behind. This is the artifact ACP, UCP and MCP all want to consume.
- AI-Powered Shopping Assistant. On your own site, where the sale usually closes, ground the assistant in that same certified data so recommendations stay on-brand and inside your approved boundaries — and so you keep the customer relationship and the checkout.
- GEO-Ready Discovery. Emit the same certified data to the discovery surfaces and engines that matter, so you are eligible to be found and represented accurately. This is readiness and eligibility, not guaranteed placement — Querytail does not control third-party AI answers, and no one can promise you a ranking or a mention.
Do this once and you are positioned for ACP, UCP and MCP together, without rebuilding every time a spec shifts. For a fuller map of how these standards are shaking out by layer, see our companion piece, the protocol wars. To see the catalog and assistant, visit the product overview.
FAQ
Do I need to implement all four protocols? No. Most of this is plumbing you consume through your platform and payment stack rather than build by hand. Your effort is best spent on the quality and machine-readability of the product data these protocols carry.
If I invest in one standard, is that wasted when another wins? Not if you invest in the data rather than the pipe. A certified, machine-readable catalog is emitted to whichever standards matter; the pipes can change without you rebuilding your product truth.
Does being "ready" guarantee my products get recommended? No. Readiness makes you eligible to be found and accurately represented; it does not guarantee rankings, mentions or traffic. AI engines decide their own answers, and Querytail is not a GEO agency that can promise placement.
Want to see what your own catalog looks like as certified, protocol-ready product truth? Book a walkthrough, or if you want to build alongside our team, apply to the Design Partner program.