WebMCP is on in your BigCommerce store. Should you leave it on?
If you sell on BigCommerce, a shopper's browser agent can now search your catalog and fill a cart on your own site, and nobody asked you first. There's a decent case for leaving it on and a narrower one for switching it off.

Most BigCommerce merchants should leave WebMCP on: the agent works inside the shopper's own browser session on your domain, sees your real prices and stock, and hands checkout back to the person. Switch it off for now if what you sell depends on rules the storefront doesn't show clearly, like B2B terms, made-to-order lead times or restricted shipping. Commerce.com turned it on by default on October 21, 2026, with an off switch in a new AI connections settings menu.
What changed on Wednesday
On October 21, 2026, Commerce.com, the company behind BigCommerce, announced with Cloudflare that WebMCP is now turned on by default for eligible BigCommerce stores. You didn't need to do anything, and you might not have noticed.
Here's what it means in plain terms. WebMCP is a proposed browser standard that lets a website tell an AI agent in the shopper's browser what it can do. Instead of the agent guessing which button is "add to cart", your store offers it tools: search products, check details, build a cart, get a checkout link. Cloudflare's example is a shopper asking for a waterproof trail shoe under $140 in size 10, and the agent making two clean calls instead of clicking around.
Three details from the announcements matter most. The tools run inside the shopper's own browser session on your domain, so the agent sees the same prices and cart the person would. The shopper still clicks checkout themselves on your store. And you can turn it off in a new settings menu that also manages all of Commerce's AI connections. If you opt out, Cloudflare says your store gets no injected script and no extra latency.
So the question isn't whether to set it up. It's whether to keep it.
The case for leaving it on
The agent sees the truth. The usual complaint about shopping agents is that they scrape pages and get prices and stock wrong. Here, the agent is asking your store directly, in the shopper's session, and getting the same answer a person would. That's better than an agent working from a stale copy of your catalog somewhere else.
The sale stays on your site. The agent fills a cart and hands back a link. The shopper finishes checkout on your storefront, with your payment options, your upsells and your email capture. That's a gentler change than agent checkouts that happen somewhere else.
It costs you nothing to run. No code, no theme change. If the browser doesn't support WebMCP, Cloudflare says the script simply exits and the page loads normally.
It's a cheap way to learn. Agent shopping is coming in some form. Walmart announced on the same day that it's opening an Agent Gateway to AI agents, with terms they must follow. Leaving WebMCP on gives you a low-risk place to watch how agent-built carts behave on your own store before the bigger platforms push harder.
Leaving it on is a decision too, so make it on purpose.
The case for switching it off
Your store has rules the page doesn't show well. The agent can only work with what your storefront tools expose. If the real terms of a sale live elsewhere, it may build a cart that looks right and isn't. Think of B2B stores with negotiated terms that depend on who's buying, made-to-order products with lead times in a PDF, or products you can't ship to certain states. Commerce.com says eligible stores include B2C and B2B, so this applies to wholesale sites too.
You can't see it. If you have no way to tell an agent-built cart from a human one in your reporting, you won't know whether it's helping or causing returns. That's not a reason to switch it off forever. It might be a reason to switch it off until you can.
Your catalog isn't ready. If product data is thin, with sizes missing or variants named inconsistently, an agent will find the gaps faster than a person would. The fix is your catalog, but the timing is yours too.
It's still early. Cloudflare describes WebMCP as an experimental proposed standard in early preview. Right now it works behind Chrome flags, in Chrome Origin Trials, in some agentic browsers and in developer tools. Very few of your shoppers can use it today, so switching it off costs you little while you get ready.
Decision guide
Leave it on, or switch it off for now
| Consideration | Leave it on if | Switch it off for now if |
|---|---|---|
| Prices | What the page shows is what the customer pays. | Prices depend on terms the page doesn't carry. |
| Catalog | Sizes, variants and details are complete. | Key answers live in photos or PDFs. |
| Shipping | Simple rules the cart already enforces. | Restrictions handled by hand after the order. |
| Reporting | You can spot agent-built carts. | You'd never know if it caused returns. |
What it doesn't change
Leaving it on doesn't mean an agent can pay for anything. Checkout stays a human click on your site.
It doesn't help agents that never visit your store. WebMCP is for agents working in the shopper's browser. Agents that shop across many sites from outside connect through BigCommerce's server-side Storefront MCP instead, which Commerce.com says offers the same set of shopping tools.
And it doesn't decide your rules for agents. Walmart's announcement is a useful reference here. It says agents must not take actions beyond what the shopper intends, can't change a customer's payment method, and can be blocked if they break those terms. Customers will be able to require approval for each agent purchase or set a spending limit. You don't need Walmart's machinery, but you do need to decide what an agent-built order is allowed to be on your store.
Our call, and what to check this week
Leaving it on is a decision too, so make it on purpose. For most stores with clear prices, tidy variants and straightforward shipping, we'd leave it on. For stores where the sale depends on terms the page doesn't carry, we'd switch it off, fix the gaps, and switch it back on.
Either way, spend half an hour on this:
- Find the setting. Open the new AI connections menu and confirm WebMCP's state. Screenshot it for whoever owns the store.
- Read your own catalog like an agent. Pick five products that cause the most support questions. Is the answer to that question in the product data, or only in a photo or a PDF?
- Check the awkward rules. Restricted shipping, minimum order quantities, customer-group pricing. Can a cart be built that breaks them?
- Decide how you'll notice. Ask whoever runs your analytics whether agent-built carts can be told apart. If not, add that to the list.
If you're on Shopify, the same thinking applies to its agentic settings, which we covered in open Shopify's Agentic page. And the gaps agents find first are usually shipping and policy answers, as in your agent checkout works, your shipping answer doesn't.
If you'd like a second pair of eyes on your catalog, cart rules and analytics before peak, that's the ecommerce work we do. Talk to us.
Related work: Paddle8.
How the work moves
A half-hour check
- Find the setting
Open the AI connections menu and record WebMCP's state.
- Read five products
Pick the ones that cause the most support questions.
- Test the awkward rules
Restricted shipping, minimums, customer-group prices.
- Decide how you'll notice
Can your analytics tell agent-built carts apart?
We build alongside the people doing the work.
Bring us the problem you’re thinking about.