Commerce.com runs almost 100,000 storefronts on Cloudflare’s network, and turned on an experimental browser standard called WebMCP across all of them without asking merchants to edit their themes or do anything by hand. Cloudflare’s own account of the rollout is worth reading end to end, because it’s one of the more concrete, non-hypothetical descriptions of what “AI agents shop on your site” actually looks like in practice, rather than the vague “get ready for the agentic future” framing that dominates most coverage of this topic.
For a PrestaShop or OpenCart store owner who doesn’t run anywhere near 100,000 storefronts, the useful question isn’t whether to adopt WebMCP specifically. It’s what this rollout, and the parallel moves happening elsewhere in commerce right now, actually reveal about what a smaller store needs to do.
What WebMCP Actually Changes
WebMCP is an experimental, early-preview browser standard from Google, currently available behind experimental flags in Chrome, in Chrome Origin Trials and in some agentic browsers, that gives an AI agent structured tools instead of a web page to interpret. Rather than an agent trying to simulate a mouse click on a search box, scroll through results, and parse a product page the way a human would, WebMCP exposes something closer to a direct function call: a tool like search_products that the agent can invoke and get a clean, structured answer back from. Cloudflare’s example is a shopper telling an agent to find a waterproof trail running shoe under $140 in size 10, and the agent executing two precise commands instead of fumbling through cookie banners and variant dropdowns the way a scripted browser automation tool historically would have.
The security model behind it is also worth understanding, because it answers the first worry most store owners have: the tool runs inside the shopper’s own active browser session, on the merchant’s own domain. No credentials get handed to a third party, and the customer keeps full control of the actual checkout step. This isn’t an agent with its own account quietly buying things on autopilot; it’s a faster way for an agent acting on a specific shopper’s behalf to read a catalog and operate the same cart that shopper sees.
Commerce.com’s Rollout Is the Instructive Part
What makes this genuinely useful to study, rather than just another vendor announcement, is how Commerce.com and Cloudflare handled scale. Cloudflare injects the bridging script at the edge, which means individual merchants on the platform didn’t need to modify their storefront code to get WebMCP support. Merchants who don’t want it can turn it off from a dashboard toggle, and Cloudflare added a zone-level setting so the script could point at Commerce.com’s own MCP server path. If a store’s MCP server is offline, or a shopper’s browser doesn’t support WebMCP yet, the system falls back gracefully rather than breaking the page.
That’s the pattern worth noticing: the infrastructure-level rollout handles the plumbing, proxying tool definitions from Commerce.com’s own backend MCP server to the shopper’s browser, so what an agent can do still depends on the catalog and cart data behind those tools. A WebMCP bridge sitting in front of a product catalog with missing variant data, vague availability information, or a return policy buried in a paragraph of marketing copy doesn’t magically fix any of that. It just gives an agent faster access to the same gaps a human shopper would also have struggled with.
A Worked Example: Prepared Versus Unprepared
It helps to picture what an agent actually encounters on two different product pages for the same item. On an unprepared page, the size, color, and material might only exist as text inside a dropdown’s visible label, stock status might read a generic “in stock” regardless of which variant is selected, and the return policy might be a sentence buried at the bottom of a separate terms page. An agent working from that page has to guess, the same way a human skimming on a phone would, whether the “in stock” label still applies to the specific size and color actually being asked about.
On a prepared page, the same three facts exist as structured data the agent can read directly: a specific SKU tied to a specific size and color combination, a stock count or availability flag scoped to that exact SKU, and a return window stated as a literal number of days with the responsible party named. Nothing about that second version requires an AI agent to exist at all; it’s just a well-built product page. The agent simply benefits from groundwork that was worth doing anyway, and that groundwork helps whichever agent protocol a shopper’s assistant ends up using.
This Isn’t Just a Cloudflare Story
Commerce.com and Cloudflare aren’t the only ones moving on this. Google announced its Universal Commerce Protocol (UCP) in January 2026, developed with Shopify, Etsy, Wayfair, Target and Walmart and endorsed by more than 20 other partners including Visa, Mastercard, Stripe, Best Buy and The Home Depot. UCP-powered checkout went live for Etsy and Wayfair inside Google’s AI Mode and the Gemini app in February, with Shopify, Target and Walmart announced as next. Separately, OpenAI built an Agentic Commerce Protocol (ACP) with Stripe and launched it as ChatGPT’s “Instant Checkout” with Etsy as the first live merchant in late 2025. Worth knowing before treating ACP as a solved checkout path: in March 2026 OpenAI scaled Instant Checkout back as a standalone feature after weak merchant uptake, with reporting citing only a small number of Shopify merchants going live, and said merchants could use their own checkout systems while ChatGPT concentrates on product discovery. None of these efforts share one standard, and a store reachable by agents from more than one AI platform should expect to stay exposed to more than one of them over time, not just whichever one a single infrastructure provider happens to support first.
That’s a genuinely different situation than most “prepare for AI shopping agents” advice acknowledges. A lot of that advice treats agentic commerce as a single checkbox to tick. In practice it’s a handful of overlapping, competing standards converging on the same underlying requirement: structured, accurate, machine-readable product and policy data, exposed through whichever specific protocol a given agent happens to speak. Store owners weighing where to host that catalog and checkout infrastructure in the first place can see what a dedicated environment looks like on our PrestaShop hosting plans page.
What a Smaller Store Should Actually Fix First
The genuinely good news buried in all of this is that readiness for agentic commerce overlaps almost entirely with things a store should already be doing for its human visitors. An agent, like a careful human shopper, needs size, color, material, compatibility, and dimensions represented as real structured fields rather than buried in a paragraph of prose. It needs a product feed where price and stock actually match what’s in the database, not what was true when the page was last manually edited. It needs a shipping and return policy stated as literal, parseable facts, “returns accepted within 30 days, buyer pays return shipping,” rather than a vague “contact us for details.”
For a PrestaShop or OpenCart store specifically, that translates into a concrete short list: check that Product and Offer structured data (schema.org markup) is actually present and accurate on top revenue pages, not just installed and forgotten years ago. Confirm stock levels sync in near real time rather than on a nightly batch job, since an agent checking availability mid-conversation won’t see tomorrow’s correction. Rewrite the returns and shipping policy page in plain, literal sentences an agent, and just as usefully a human skimming on a phone, can parse without inference.
What to Leave Alone for Now
It’s just as useful to be clear about what doesn’t need immediate action. A store doesn’t need its own WebMCP server, its own ACP integration, and its own UCP integration all built in parallel this quarter. The standards themselves are still settling, adoption is concentrated among large platforms and their flagship merchants, and the infrastructure-level rollouts like Commerce.com’s suggest that a meaningful share of this will eventually arrive the way WebMCP did there: as a platform-level capability a host or CDN switches on, not a from-scratch build every individual merchant has to do alone.
The realistic sequencing for a small store is to fix the data hygiene now, because that work pays off for human shoppers immediately and for agents whenever they show up in meaningful volume, and to treat the specific protocol integrations as a decision to revisit once actual purchase volume from AI agents, rather than hypothetical future volume, shows up in analytics. Chasing every emerging standard the week it’s announced is a good way to spend a quarter building integrations for traffic that hasn’t arrived yet.