Frequently asked questions

It sits between an AI assistant and your Shopify store. The assistant cannot write to your store at all — it can only propose a change. The app reads your store, shows you exactly what would change, and applies it only when you approve.

Claude, through Claude’s connectors. It uses the open Model Context Protocol (MCP), so other assistants that can connect to a remote MCP server with sign-in should work too — but Claude is the one it is built and tested with.

After installing, the app’s home screen shows a server address. Add it to Claude as a connector. Claude opens a page asking which store, and you approve the connection inside your own Shopify admin. There are no API keys to copy and nothing to install on your computer.

No. Claude is never given a tool that writes to your store. Every change waits in your approval queue until you — or an auto-approval rule you set up for low-risk edits on a paid plan — approve it. Claude cannot approve its own proposals.

The app re-reads your store before writing. If anything the change touches has moved since it was proposed — a manual edit, another app, an earlier change — approval is refused and nothing is written. The review page warns you before you click, not after.

Almost always, in one click. Edits are fully reversible because the app keeps the previous value of every field it writes. Creating something can be undone by removing it again. Deleting a product cannot be truly undone — re-creating it gives it a new identity, so old orders and links no longer point at it — and the app says so before you approve, not after.

Low risk is unlikely to change what customers see or pay. Medium is visible to customers or reorganises the store, such as a title, an SEO description or a menu. High risk touches prices, availability or discounts, or deletes something. High-risk changes always need a person and can never be auto-approved.

It arrives as one job you approve once, graded as risky as its riskiest item — so a price change can’t hide among SEO edits. It applies one item at a time with progress, and if a few fail the rest still land and the failures are named.

Products, their variants, stock levels and images, collections (including which products they hold and the rules of automatic ones), pages, blogs and articles, navigation menus, URL redirects, metafields (and their definitions) and metaobjects, translations (in any of your languages, published or not, and for a single market where wording differs), discounts, and the name, status and price adjustment of markets. You can switch any of these off in the app’s settings.

Orders and payment details — the app holds no permission for them, so no request can reach them however it is worded — and individual customers. It does hold read access to customers, because that is the only way Shopify lists customer segments, and it uses it for segment names and ids alone: it never reads, stores or changes a customer’s details.

Yes. It can add products to a manual collection or take them out, and change the rules of an automatic one — tag, title, type, vendor and price. A rule change is high risk, because it moves every matching product at once. It will not hand-pick products into an automatic collection, or add rules to a manual one that already has products: mixing the two makes the collection unreadable to other apps on older versions of Shopify’s API. Collections built with Shopify’s newer features are left to you in the admin.

Create percentage, fixed-amount and free-shipping discounts, as a code or automatic, and change their value, the products or collections they apply to, minimum purchase, combinations, usage limits, dates and status, and limit it to customer segments or open it to everyone. Every discount change is high risk, so it always needs a person and can never be auto-approved. A new discount must say when it starts, so you see when it goes live before you approve it, and a discount over half price or one that more than doubles is flagged for a second look.

Aim a discount at named customers — it can limit one to customer segments, by name, but never to individual people; change an existing code, because copies of it are already out there; edit buy-X-get-Y discounts beyond their title, dates and status; or delete a discount — switching it off has the same effect and can be undone.

Claude proposes the number to set at each location, and the review shows the current number beside it. If a sale moves the stock before you approve, nothing is written and the change says why — the app never sets stock from a number that is out of date. For the same reason, undoing a stock change only works until the next sale.

Yes, but it is always high risk and the review says plainly that it cannot be undone: a re-created variant is a new one, and its orders and stock history stay with the old. Removing an image can be undone, because the file is kept.

Its name, whether it is active, and the percentage its prices sit above or below your base prices — always graded high risk, apart from a rename. Countries, currencies and domains stay with you: moving a country into one market can take it out of another, and that is not something to approve from one row. If a market shares its price list with another, the app refuses to change it rather than re-price both.

When you connect Claude, you are allowing it to look things up in your store so it can propose sensible changes. What it reads goes to Claude’s operator under their privacy policy. Our privacy policy sets out exactly what this app stores.

Access to your store ends immediately. Your change history is kept for 48 hours in case you reinstall, then everything the app holds for your store is permanently deleted.