Key takeaways
ORI’s 30-second summary
- A growing share of online purchases now starts with an AI assistant, not a search box or a homepage.
- MCP lets agents read your catalogue; ACP and UCP standardise how they check out and pay.
- Clean structured product data, real-time stock and pricing, and a reliable API matter more than visual polish for this channel.
- Start with structured data and a read-only product feed, then add agent checkout once the basics are solid.
For twenty years, online retail has been designed for a person looking at a screen. Product pages, filters, banners and checkout flows all assume a human is clicking. In 2026 that assumption is breaking. Shoppers increasingly ask an AI assistant to "find a waterproof jacket under $150 that arrives by Friday" — and the assistant does the browsing, comparing and, more and more often, the buying.
This is agentic commerce. Shopify has reported AI-attributed orders growing roughly elevenfold between January 2025 and January 2026, and almost every major platform — Google, OpenAI, Stripe, Amazon and Walmart among them — has shipped a standard for how agents discover products and pay for them.
The three protocols, in plain English
The landscape looks confusing because several standards arrived at once. In practice they work at different layers and complement each other.
MCP — Model Context Protocol
MCP is the general-purpose connector between AI assistants and outside systems. Exposing an MCP server lets an agent search your catalogue, read product details, check stock and look up order status in a structured way, instead of scraping your pages.
ACP — Agentic Commerce Protocol
ACP standardises the checkout itself: how an agent builds a cart, confirms the price, passes a secure payment token and receives an order confirmation, while the merchant stays the seller of record.
UCP — Universal Commerce Protocol
Announced by Google in early 2026, UCP brings native checkout, loyalty and member pricing into AI surfaces such as AI Mode in Search and the Gemini app.
You do not have to pick a winner today. Most stores should start with clean data and an MCP-style product feed — that groundwork is shared by every protocol.
Why agents see your store differently
An agent does not care about your hero banner. It cares about whether it can answer the shopper's question accurately and complete the purchase without errors. That changes what "good" looks like:
- Complete attributes: size, material, dimensions, compatibility and care instructions as structured fields, not buried in prose.
- Real-time truth: stock levels, delivery estimates and prices that match what checkout will actually charge.
- Clear policies: returns, warranty and shipping rules that a machine can read and quote back.
- Reliable APIs: fast, documented endpoints with sensible rate limits and predictable errors.
If an agent finds conflicting prices or vague stock information, it will simply recommend a competitor whose data it can trust.
A practical readiness checklist
Step 1: Fix your product data
Audit your catalogue for missing attributes, inconsistent units and duplicate listings. Add complete Product, Offer and MerchantReturnPolicy structured data to every product page. This also helps traditional search and AI answer engines.
Step 2: Expose a read-only product feed
Publish an authenticated feed or MCP server that lets assistants search products, read details and check availability. Keep it read-only at first — it carries little risk and teaches you how agents use your data.
Step 3: Prepare checkout for tokens, not forms
Agent checkout relies on delegated payment tokens and server-to-server confirmation rather than a customer typing card details. Make sure your order and payment services can accept an order through an API with the same fraud checks, tax rules and confirmations as your web checkout.
Step 4: Measure the channel
Tag orders that originate from agents and assistants. Track conversion, return rates and average order value separately, so you can see whether the channel is profitable before investing further.
Risks to plan for
- Fraud and abuse: agents can place many orders quickly, so rate limits and anomaly detection matter.
- Price accuracy: a stale feed can create orders at prices you did not intend to honour.
- Brand distance: the assistant, not your website, now frames the first impression of your product.
The bottom line
Agentic commerce will not replace your website, but it is quickly becoming a second storefront that you do not design. The merchants who win there will be the ones with the cleanest data and the most dependable APIs. The good news is that this work also improves search visibility, marketplace listings and your own site's filters — so it pays off even before agents become a large share of your revenue.