On September 28, TechCrunch reported that Shopify has extended support for the Web Model Context Protocol so browser-based AI agents can complete a purchase, not just browse. Agents could already look at products and fill a basket. Now there are three new tools, named plainly: get checkout, update checkout, complete checkout. An agent can inspect the checkout, change the delivery option or address, and submit the order once the buyer authorizes it. No screenshots, no scraping.
It is rolling out to eligible Shopify merchants and runs on WebMCP alongside Shopify’s Universal Commerce Protocol, with Shop Pay supported. The Muse agent from Meta is named among the agents it works with.
Shopify staff product manager for agentic commerce, Gil Greenberg, put the motivation this way: “Shopping with an agent shouldn’t feel like watching paint dry.”
The part that matters to you
If you sell online, the person buying from you may soon never see your shop.
Think about what that removes. The photography. The layout. The carefully written paragraph above the add-to-basket button. The trust you were building with a well-made page. An agent comparing your product against three others does not read your page the way a customer does. It reads your data.
So the things you may have treated as admin become the sales pitch. Product titles that describe the thing rather than tease it. Variant names that are not internal shorthand. Accurate stock. Delivery options that are real and priced. A returns policy written where a machine can find it, not buried in a PDF. Fill in specifications completely, including the fields you have been skipping since 2019.
This is not a new kind of work. It is the same discipline that made your listings rank, applied to a reader with no patience and no eyes.
Do not throw away the brand work.
A tempting conclusion follows, and it is wrong. The conclusion is that brand no longer matters because a machine is doing the buying.
Two things argue against it. First, the buyer still authorizes the purchase and still decides which shop they would rather deal with when two options are nearly identical in the data. Second, the agent itself decides what to recommend, based on how the wider web describes you. Reviews, mentions, consistency of name and claim across every place you appear. A shop with a coherent story and a clean reputation gets recommended. A shop with four versions of its own name does not.
The brand work does not disappear. It moves earlier in the chain, where it has always done its best work and received the least credit. The practical end of that is making sure the AI systems doing the recommending can recognize and describe your business correctly, which is a different job from ranking a page.
What to do in the next fortnight
Pull your product feed and read it as a stranger would. Find the fields that are empty, the titles that mean nothing out of context, and the prices that are wrong somewhere. Fix those before you touch anything clever.
Then check your own name. Search for your business and see what comes back. If a machine is going to summarize you to a buyer, the summary is assembled from whatever is already lying about.
Figuring out what your business is actually called, what it claims, and whether the claim holds up everywhere it appears is the foundation for everything else. That is the argument of Breaking Boundaries, and agentic checkout has just raised the cost of getting it wrong.