The Short Answer
For most Magento stores, enable UCP first, then read-only MCP, then ACP, and treat WebMCP as a pilot. UCP is where AI checkout is live at scale today, in Google AI Mode and Gemini. MCP is a low-risk way to let agents read your catalog. ACP now mostly feeds ChatGPT product discovery. WebMCP is still a draft.
That order assumes you sell to US shoppers and use Google Merchant Center. If your traffic data says otherwise, reorder it. The section on choosing shows how.
Key takeaways
• UCP (Universal Commerce Protocol) comes from Google, built with Shopify and big retailers. It powers checkout in Google AI Mode and Gemini for eligible US merchants.
• ACP (Agentic Commerce Protocol) comes from OpenAI and Stripe. ChatGPT’s Instant Checkout was scaled back on March 24, 2026, but the protocol is still being developed.
• MCP (Model Context Protocol) is a general way for AI agents to use tools. It is now run by the Linux Foundation.
• WebMCP is a draft web standard that lets a web page offer tools to an AI agent inside the browser. It is not yet a W3C standard.
• Fix your product data first. Every protocol is only as good as the catalog behind it.
What Is Agentic Commerce?
Agentic commerce is when an AI assistant helps a person find a product, compare options, and, in some cases, make a purchase. Instead of opening ten browser tabs, the person asks a question, and the assistant does the work.
For that to work, the assistant needs to know your store. It has to understand what you sell, how much it costs, if it’s in stock, and how someone can buy it. Protocols are the rules that let the assistant ask questions and the store answer them, without needing custom setup every time. Think of them like menus and order forms. They’re the agreed-upon way that AI assistants and shops talk to each other.
The Four Protocols at a Glance
|
Protocol |
Who is behind it |
What it does |
Status (as of late October 2026) |
Effort for Magento |
|
UCP |
Google, with Shopify, Etsy, Wayfair, Target, Walmart |
Let's AI experiences find your merchant, read your catalog, and run checkout |
Live in Google AI Mode, Gemini and Search for eligible US merchants. Canada and Australia announced, UK later |
Medium |
|
ACP |
OpenAI and Stripe |
Product feeds, carts and orders between ChatGPT and merchants |
Instant Checkout scaled back March 24, 2026. Spec updated April 17, 2026. ChatGPT now sends shoppers to merchant checkouts. Still marked beta |
Medium |
|
MCP |
Created by Anthropic, now under the Linux Foundation |
Lets agents call tools, such as search catalog or create cart |
Mature and widely supported. Donated to the Agentic AI Foundation on December 9, 2025 |
Low to medium |
|
WebMCP |
W3C community group, led by Google and Microsoft engineers |
Lets a web page expose tools to an agent in the shopper’s browser |
Draft report, not a standard. Chrome origin trial from version 149. API names still changing |
Low, but moving target |
What Is UCP?
UCP, the Universal Commerce Protocol, is Google’s open standard for shopping with AI. Announced at NRF on January 11, 2026, it covers the whole path from finding a product to checking out in Google AI Mode and Gemini.
Here is what matters for a merchant. Google says UCP checkout is available to eligible US merchants. Reports describe a Buy button reaching main Google Search results in May 2026, with checkout expanding to Canada and Australia, and the UK later. Merchants usually get in through Google Merchant Center plus a UCP integration or waitlist. The retailer stays the merchant of record.
A Magento store typically takes part by publishing a UCP merchant profile at /.well-known/ucp, keeping a healthy Merchant Center feed, and exposing catalog and cart capabilities. UCP is the only one of the four with live, at-scale checkout today, which is why it leads the ranking.
What Is ACP, and Is It Still Worth Doing?
ACP, the Agentic Commerce Protocol, is the standard OpenAI and Stripe launched on September 29, 2025, to power Instant Checkout inside ChatGPT. OpenAI scaled that feature back on March 24, 2026, so ChatGPT now sends shoppers to merchants’ own checkouts.
The protocol did not disappear. A release dated April 17, 2026 added carts, a Feed API, orders, acp.json discovery and an MCP binding, according to Anatta. In a statement reported by Digital Commerce 360, OpenAI said it is focusing on search and product discovery, with ACP connecting people to merchants. Reports also say few merchants ever went live on Instant Checkout.
So ACP is still worth doing, but for a different reason than a year ago. It is mainly about getting your products into ChatGPT discovery, not about buying inside the chat. If you use Stripe, or ChatGPT traffic matters to you, it moves up the list.
What Is MCP?
MCP, the Model Context Protocol, is an open standard that lets AI applications connect to outside tools and data. Anthropic introduced it in November 2024, and on December 9, 2025, donated it to the Agentic AI Foundation, a Linux Foundation fund co-founded with Block and OpenAI.
In commerce, an MCP server is a doorway into your store for any compatible agent. A store might offer tools like search_catalog, get_product, and create_cart. Unlike UCP and ACP, MCP is not about shopping specifically. It is general plumbing, which is both its strength (works with many agents) and its weakness (no built-in idea of a merchant, checkout, or payment).
The practical advice: start with read-only tools. They let agents look at products without touching customer, order, or payment data, and they carry the lowest risk.
Disclosure: MCP was created by Anthropic, the company that makes the assistant that helped draft this article. The ranking here rests on the status evidence above, not on that history.
What Is WebMCP?
WebMCP is a proposed web standard that lets a web page offer tools to an AI agent running in the shopper’s browser. The W3C Web Machine Learning Community Group developed it, with editors from Google and Microsoft, and it remains a Draft Community Group Report as of October 2026.
It is early. Chrome announced an early preview on February 10, 2026, and now offers an origin trial from version 149. The API has already changed: navigator.modelContext is deprecated in favor of document.modelContext, per Chrome’s documentation as summarized in October 2026. Other browsers are still watching. Building for it now means rebuilding later, so use it as a small experiment, not a foundation.
So Which Should You Enable First?
Choose based on where your shoppers come from, not which protocol is newest. Here is the ranking for a typical Magento store selling to US shoppers, and what would change it.
|
Order |
Protocol |
Why |
What would change it |
|
1 |
UCP |
The only live, at-scale agent checkout, with Google surfaces that already send shoppers |
You do not sell in UCP markets, or Google brings you little traffic |
|
2 |
MCP, read-only |
Low risk, works with many agents, and the same catalog tools can power internal and partner use |
You have no developers or agencies to test and monitor it |
|
3 |
ACP |
Gets products into ChatGPT discovery. Checkout now happens on your site |
Your analytics show real ChatGPT referrals, or you already use Stripe, which moves it up |
|
4 |
WebMCP |
Interesting, but still a draft with changing APIs |
A Chrome-heavy audience and a team that enjoys experiments |
Let your own traffic decide
Google Analytics introduced an AI Assistant feature on May 13, 2026, so you can see how much traffic comes from AI tools, as Google’s release notes explain. Check it before you decide. If traffic from Gemini or Google AI Mode is important, UCP should be the choice. If ChatGPT referrals dominate, move ACP up. If you see almost none from either, focus on product data and revisit in a quarter. Our updated GA4 with GTM guide shows how to set up the reporting.
Before Any Protocol: Fix the Basics
A protocol can only share what your catalog has. If the data is weak, the answers will be weak too. Agents are less forgiving than shoppers because they compare stores at the same time.
• Titles and descriptions that say what the product is. Plain nouns, key attributes, no keyword stuffing. AI Content Generator can help draft missing copy, but review it.
• Complete attributes. Size, color, material, compatibility, brand, and identifiers such as GTIN. Cleaner filters help too, as our navigation guide explains.
• Accurate price and stock. If an agent promises a price or availability you cannot honor, you lose the sale and the trust.
• Clear shipping and returns. Agents compare these. Put them in one findable place.
• Healthy feeds. Fix Merchant Center errors first. MageDelight’s Product Feeds can help produce clean feeds.
• Honest reviews. Agents lean on review signals. Keep yours genuine and balanced, as covered in our FTC review rule guide.
• Several languages? Keep translated catalogs consistent. See the translation workflow guide.
Where MageDelight's Agentic Commerce Suite Fits
MageDelight’s Agentic Commerce Suite connects a Magento 2 store to compatible AI agents through MCP, UCP, ACP, and WebMCP. According to the product page, Magento stays the source of truth for catalog, pricing, customers, orders, cart, and checkout, and when enabled, the suite can prepare shopping carts for customers. The meta description also mentions llms.txt, catalog APIs, agent carts, and analytics.
The page lists first-year prices of $149 for Magento Open Source, $299 for Adobe Commerce and $349 for Adobe Commerce Cloud, then $99 a year for updates and support. A demo store and a scheduling link are available. Because this is a four-protocol suite, you can turn on the pieces in the order above instead of buying four tools.
The writer had only the product page, not a user manual, so confirm these points with MageDelight before you buy:
• How deep each protocol goes. Does UCP include checkout, or only the merchant profile and catalog?
• Whether each protocol has its own on and off switch.
• Which products and categories you can exclude, such as age-restricted, made-to-order, or gift items.
• What the analytics show, and whether they separate each agent.
• How read-only surfaces are protected, including rate limits and data exposure.
• Which Magento versions and themes are supported, including Hyvä.
Do not confuse it with MageDelight’s AI Agent MCP Integration, which is aimed at managing a store with AI, not at shoppers’ agents. The LLMs.txt Generator is a separate, smaller tool for guiding AI crawlers.
A Safe Rollout in Seven Steps
1. Take a baseline. Note your AI Assistant channel traffic in GA4 and your Merchant Center feed health today.
2. Fix the data basics from the section above, starting with your top 50 products.
3. Install on staging. Do not test on live data first.
4. Turn on read-only surfaces first: the UCP profile and read-only MCP tools. Confirm they expose no customer, order or payment data.
5. Decide what agents may do with each product. Keep cart and checkout off for items that need human judgment, such as age-restricted goods, custom orders, and anything with legal limits.
6. Test with real agents. Ask Gemini, ChatGPT, and Claude to find, compare, and add a product to a cart, and note every wrong price, missing size, or odd answer.
7. Go live in stages, watch the numbers weekly, and keep a one-click way to switch each surface off.
Risks and Unknowns
• Standards are still moving. ACP changed direction in March 2026, and WebMCP changed its API name this year. Expect updates.
• You stay responsible. The merchant remains the seller. Wrong prices, stock errors, and returns are still yours to handle.
• Fraud and abuse. New doors attract bots. Rate limit, monitor carts, and keep payment on your own checkout.
• Data exposure. Keep read surfaces limited to public catalog data.
• Reach is uneven. UCP checkout is US-first today. Check eligibility for your markets.
• Unproven demand. Reports say few merchants went live on Instant Checkout. Start small and measure.
Common Mistakes
• Enabling everything at once. You will not know what worked or broke.
• Skipping product data. The protocol is not the bottleneck. The catalog is.
• Assuming ChatGPT checkout still exists. It was scaled back in March 2026.
• Building deeply on WebMCP. The API is still changing.
• Opening checkout to every product. Exclude items that need a human.
• Never check your analytics. Let the AI Assistant channel guide you.
Which Setup Fits Your Store
• US store, Google Merchant Center, steady Google traffic: UCP first, then read-only MCP.
• Stripe user with visible ChatGPT referrals: UCP and ACP feeds together, then MCP.
• Outside the UCP markets: start with data, feeds, and read-only MCP, and watch for expansion.
• Small catalog, limited team: do the data work and one protocol well.
• Agency or developer-led store: MCP gives you a flexible base for experiments.
• Need help? MageDelight’s Technical SEO Audit, Digital Marketing Audit, and support service can help prepare the store, and Hyvä development covers front-end work.
FAQ
What is the difference between UCP and ACP?
UCP is Google’s protocol for shopping and checkout in Google AI Mode and Gemini. ACP is OpenAI and Stripe’s protocol, built for ChatGPT, which now focuses on product discovery and sends shoppers to the merchant’s own checkout.
Is ChatGPT Instant Checkout still available?
OpenAI scaled it back on March 24, 2026, and moved purchases toward merchant apps and merchant checkouts. The ACP specification is still being developed, with an update dated April 17, 2026.
Is MCP the same as UCP?
No. MCP is a general way for AI agents to use tools. UCP is specific to commerce. They can work together, since a UCP checkout can also be offered as MCP tools.
Should I build for WebMCP now?
Only as an experiment. It is a W3C community draft; Chrome offers an origin trial. The API name has already changed from navigator.modelContext to document.
Do I need all four protocols?
No. Start with the one that matches your traffic, UCP. Then add read-only MCP. Add ACP and WebMCP when your data or audience justifies it.
Will turning these on expose customer data?
It should not. Read surfaces should show public catalog data. Payment should stay on your checkout. Confirm this with your vendor. Test it on staging.
How do I know whether AI traffic matters to my store?
Check the AI Assistant channel in GA4, which Google added on May 13, 2026. Compare the AI Assistant channel's conversion rate and revenue per session with those of search.



