Ask Claude to rewrite your SEO, fix a menu or put a range on sale, in plain English. It proposes; you see exactly what would change, field by field; nothing reaches your store until you say so.
Performance Running Shorts
PendingHigh riskFully reversible“Summer sale on two sizes — 20% off, with the old price shown as compare-at.”
| Field | Currently | Would become |
|---|---|---|
| Black / S — Price | $29.99 | $24.99 |
| Black / S — Compare-at price | Not set | $29.99 |
| Black / M — Price | $29.99 | $24.99 |
Nothing is written until you approve.
0 changes
reach your store without passing your approval queue. Claude is never given a tool that writes to your store — it can only propose. That is true by construction, not by policy.
Four steps, and only one of them writes to your store.
Add the app from the Shopify App Store, then connect Claude from its setup screen. You approve the connection inside your own Shopify admin — no terminal, no API keys.
“Write SEO descriptions for the products that don’t have one.” “Put the winter coats on sale.” Claude looks things up and queues its proposals.
Each change shows every field’s current value beside what it would become, graded by risk, and checked against your live store.
Approve and it is applied. Reject with a reason and Claude reads it before proposing anything similar. Changed your mind? Undo in one click.
The review step is the product, so everything around it is designed not to let you down.
Every proposal is read against your live store first, so you review the actual before and after of every field.
Prices, availability and deletions are always high risk and always need a person — they are never auto-approved.
If something changed since Claude proposed, approval is refused and nothing is written. The review page warns you before you click.
Applied changes revert in one click. Anything that cannot be undone says so up front, not afterwards.
A hundred product edits arrive as one job, graded as risky as its riskiest item, applied with progress.
Who proposed it, who approved it, when, and what it replaced — kept for as long as you have the app.
Thirteen kinds of store content — each one you can switch off in the app’s settings.
Titles, descriptions, SEO and tags; prices, SKUs, stock per location and new or removed variants; images and alt text.
Titles, descriptions, sort order and SEO; products in manual collections and rules in automatic ones.
Content, titles and URLs.
Add, rename, reorder and remove items — removals are flagged, never silent.
Point old addresses somewhere useful.
Custom data such as care instructions, fit notes and size guides — including creating the field itself.
Translated titles and content in any of your languages — including ones you have not published yet — and wording for a single market.
Create and adjust value, products, minimums, dates and customer segments — always high risk, never auto-approved. Never aimed at named customers.
Name, status and percentage price adjustment. Countries and currencies stay with you.
Deliberately out of reach: orders, payment details and individual customers. The app holds no permission for orders or payments, and reads customer data only to list your customer segments by name — so no request, however it is worded, can touch a customer’s details.
Review, the before-and-after diff, the audit trail and one-click undo are on every plan. Paid adds unlimited jobs and auto-approval for low-risk edits.