AI agents can write code and pass the bar exam, but buying shoes online remains difficult. Many agents have to interpret a website’s HTML, guess which buttons to click and recover when they make mistakes. A theme change can disrupt that process.
Cloudflare and Commerce.com are testing another approach: give browser-based agents structured shopping tools, such as product search, instead of making them navigate a page like a person. The companies’ WebMCP storefront preview shows how this could work at scale, while keeping shoppers on the merchant’s site.
For business owners, this is a development to watch, not a switch every store should rush to flip. WebMCP is still an experimental proposed browser standard in early preview. But its design raises useful questions about how your product information, checkout and customer control should work when an AI agent is involved.
What WebMCP changes
Today, many shopping agents work from a webpage’s HTML. They inspect what is visible, try to identify the right controls and simulate clicks. Cookie banners, variant menus and bot challenges can make this unreliable. The process can also be slow and use many tokens.
WebMCP gives agents structured tools on a page. Instead of hunting for a search box and guessing how to use it, an agent can discover and call a tool such as `search_products`. The store can offer tools for finding products, checking details, building a cart and getting a checkout link.
In Cloudflare’s example, a shopper asks an agent to find a waterproof trail running shoe under $140 in size 10 and add it to the cart. The agent calls a product-search tool with those criteria. Because it runs in the shopper’s active browser session on the store’s domain, the cart updates there. The shopper still clicks checkout themselves.
That last step matters. The example keeps the shopper in control of completing the purchase and their payment details. WebMCP helps an agent interact with a storefront; it does not mean the agent must take over the entire transaction.
Why the Commerce.com example matters
Commerce.com wanted to enable WebMCP across almost 100,000 storefronts using Cloudflare. Its stores run on shared zones and can use merchant vanity domains. Turning the feature on for a whole domain zone would not, by itself, give each merchant a simple way to opt out.
The companies connected WebMCP to Cloudflare Configuration Rules and custom hostname metadata. Commerce.com can enable the feature across its main zone, while a merchant can switch it off in the Commerce.com dashboard. Cloudflare then bypasses WebMCP for that storefront at the edge. An opted-out store gets no injected script, remote lookup or added latency from WebMCP.
This is a practical lesson for any platform or business considering agent access: a shared rollout needs a clear way to respect individual store choices. If your site is part of a larger platform, find out who controls the feature and how your store can decline it. If you operate your own store, make sure any integration has a defined scope and an off switch.
Browser agents and server-connected agents are different
WebMCP supports agents working in a shopper’s browser. Commerce.com’s server-side MCP supports agents that connect directly to a store, such as an assistant helping someone shop across several sites. The companies say both approaches can use the same set of shopping tools.
That distinction affects how you plan. A browser-based agent works in the active session and sees the same prices and cart as the shopper. In the described integration, Cloudflare serves as an edge delivery layer for Commerce.com’s backend MCP server, proxying tool definitions and actions unchanged. Merchants retain control of their tool logic.
These are details of this integration, not a promise that every agent or commerce platform works the same way. Before adopting a tool, ask where it runs, what data it can access, who maintains its actions and what happens when the service is unavailable. Businesses exploring tool-using AI agents should define permitted actions and approval points before connecting an agent to customer-facing workflows.
What businesses should not assume
WebMCP is an experimental proposed browser standard in early preview, not a universal feature that works in every browser. The source says it is available behind experimental Chrome flags, in Chrome Origin Trials, in some agentic browsers and in developer environments such as Cloudflare Kitesurf. Availability may change as the standard evolves.
Nor does adding tools automatically make a store ready for agent-led shopping. The tools need to reflect accurate product details and the store’s real commerce flow. Agents must also handle unavailable products, variations and errors sensibly. A useful starting point is to review whether your product catalogue is complete and consistent across your website and sales channels. If you are improving product content, an AI product description generator may help with titles, descriptions, feature lists and metadata.
Keep the customer experience in view, too. The goal in the Commerce.com example is for the agent to help with product discovery and cart building while the shopper remains on the merchant’s site and chooses when to check out. Decide which steps an agent may perform and which should require a person’s confirmation.
What to do next
You do not need to rebuild your store because of this preview. Use it to prepare a practical checklist and speak with your platform provider about the capabilities and controls it supports.
- Ask your provider about support and availability. Find out whether it supports WebMCP, server-side MCP, or neither. Confirm whether the feature is experimental and which browsers or environments can use it.
- Check what an agent would be allowed to do. List the tools you would consider useful, such as searching products or building a cart. Decide where human approval is needed, especially before checkout.
- Confirm control and privacy arrangements. Ask how an individual store can opt out, what session information an agent can access and whether credentials are shared with a third party. The Commerce.com integration says its browser tools run on the store’s domain and no credentials go to a third party; verify the details for your own setup.
- Review your catalogue and shopping flow. Check product names, options, prices and availability. Make sure your checkout instructions and customer support process are clear. A website audit can help you identify website issues to address alongside this work.
- Test carefully before expanding access. If your platform offers a safe test environment, try ordinary product searches, different product options and an unavailable service. Confirm that the storefront still loads normally if WebMCP is unsupported or its MCP server is offline. The Commerce.com integration is designed to fall back in those cases.
The immediate opportunity is not to hand your whole shop to an agent. It is to make your products and approved shopping actions easier for new interfaces to understand, while keeping control of the customer relationship and the final purchase.
Sources
- How Commerce.com and Cloudflare are getting storefronts ready for AI agents — Cloudflare Blog, Oct 21, 2026
This article was drafted with AI from the sources above, then published only after passing 20 automated checks: every figure must appear in the sources, links must work, and nothing may be invented. Spotted an error? Tell us and we'll correct it.
Want us to do this for you?
We help businesses across 3 continents implement everything you just read — and a lot more.