On October 21, Commerce.com and Cloudflare switched on WebMCP across almost 100,000 storefront domains. Eligible BigCommerce stores got it by default, and merchants can opt out in the Control Panel. An AI agent working in a shopper’s browser can now search the catalog, check product details and build a cart through structured tools instead of guessing which button to click.
Shopify got there first. Its developer changelog from August 5 says WebMCP tools are live on every Liquid storefront, with nothing to install or configure.
That means two platforms that many mid-market brands run on have turned on an agent-facing layer by default. If you sell on either, you didn’t have to do anything to get it. That’s the point, and it’s also the risk.
What actually changed
WebMCP is a proposed browser standard, still experimental, that lets a page register tools an AI agent can call. Shopify notes that agent support is currently limited to Chromium-based browsers through an origin trial, which caps how many shoppers can use it today. The platforms shipped the plumbing ahead of the demand.
The details matter more than the headline. The agent runs inside the shopper’s live session on your domain. Cloudflare’s write-up is explicit that the shopper still clicks checkout. Shopify’s changelog adds the line every eCommerce team should read twice: “Cart tools call the same standard storefront actions that apps use, so if your theme opens a cart drawer on updates, agents trigger it too.”
So agents don’t get a separate, clean channel. They drive your storefront: your theme, your cart drawer and every app hooked into it.
Why this is a performance question
Agents don’t wait. Stripe’s technical field guide to agentic commerce (March 10, 2026) says agents run on strict timeout budgets: if your API takes two seconds to answer an inventory check, the agent may time out or hallucinate a failure.
Now picture what happens when an agent adds a product to the cart. The theme opens the drawer. The free-shipping bar recalculates. The upsell app fetches recommendations. Analytics and ad pixels fire on the cart event. That’s the same queue of work a human shopper waits through, and it’s measured as interaction latency. Shopify’s analysis of its own stores (April 27, 2026) found that 32 milliseconds of interaction lag costs 1.5% of conversion. The agent now inherits that queue, and is much less likely to be tolerant of performance issues.
On BigCommerce there’s one more piece: a script. Cloudflare injects a small bridge script at the edge to expose the tools. Cloudflare says it exits cleanly when a browser doesn’t support WebMCP, and opted-out stores get no injected script at all. That’s well designed. It’s also one more script on your pages that nobody on your team added, which is how script inventories drift. Third-party apps already account for 44% of eCommerce load time, based on Yottaa’s data, and this is one more variable added to that equation.
Discovery moves. The sale still happens on your site.
It’s tempting to read agentic commerce as the moment the storefront stops mattering. But both rollouts point the other way. The agent finds the product and builds the cart, then hands the shopper back to your checkout.
The early numbers on off-site buying aren’t encouraging either. The commercetools Agentic Commerce Radar (April 22, 2026) reported that Walmart saw three times lower conversion in chat than on its own website. Discovery is spreading across agents and assistants, but the transaction keeps coming home, making it more important than ever that your PDP and checkout experiences are running smoothly.
What to do before peak
Black Friday is November 27. Four things are worth doing now:
- Make the WebMCP decision on purpose. On BigCommerce, check the MCP settings in the Control Panel and decide whether to stay on. On Shopify, know that it’s on and test it.
- Re-inventory scripts on product and cart pages. Compare the hosts loaded on a logged-out product page with the apps you think you run, and include anything the platform or CDN injects.
- Measure cart interactions, not just page load. Track INP on add-to-cart and the cart drawer in field data, by template and device. Move non-essential cart listeners, like upsells and secondary analytics, behind the cart update itself.
- Plan for messier attribution. In-browser agents act inside a real shopper’s session. Ask your analytics and monitoring vendors how they will tell agent-assisted sessions apart before you trust a holiday conversion report.
Where Yottaa fits
This is the layer Rapid works on. Yottaa’s Audit Third-Party Tags inventories every third-party script on the site and benchmarks each one with a Performance Impact Rating. Application Sequencing sets the load order so critical content and interactions come first and non-essential tools load after, without code changes. Yottaa RUM shows field LCP and INP by page type, so you can see whether the cart is getting slower as apps and platform features pile up.
Ready for agents means fast for people
With browser support where it is, few of your shoppers will use a WebMCP agent this holiday season. But the platforms have decided agents are coming, and they’ll use the same storefront your shoppers do. The fastest way to be ready for them is to make that storefront fast for people.
Has your team decided whether to leave WebMCP on?
Subscribe to our blog
Performant Experimentation – A Guide to A/B Testing
Every A/B test you run is a purchase. You’re buying information—an answer to a hypothesis—and you’re paying for it in degraded performance and lost conversion during the test window.
Download eBook
