Connect Claude to Shopify: What It Can and Can't See

Connecting Claude to Shopify lets it read your sales, orders, products, customers and inventory, and, if you approve write access, create products, set stock and make discount codes. It sees only Shopify's data. Ad spend from Meta or Google, what shipping cost you and the cost of handling returns sit outside it, so its profit answers leave those costs out.

That last point is where owners get caught. Ask "which product makes the most money?" and Claude will answer confidently from what it can reach: sales minus whatever product cost you typed into Shopify. A cushion with $12,600 of sales can look like your best line and be a loser once the adverts that sold it are counted. Below: the setup, what each kind of question can and cannot return, and how to fill the gaps.

Follow me on Instagram@sagnikteaches

What the connector is and how it gets into your store

The Shopify connector is an official one: Claude's connector directory lists Shopify as the maker and marks it as verified by Anthropic. It works through the Model Context Protocol, the standard that lets an AI chat call outside tools. In plain terms, Claude gets a menu of actions it can ask your store to perform. The listing counts 25 tools, with names such as search products, list orders, get inventory levels, run analytics query, create product, set inventory and create discount, plus a switch-shop tool and a few that send raw queries to Shopify's data interface. Shopify's own help page for the connector lists a longer set of action names, including theme editing and digital products, and each action's name appears in the chat as it runs, so you can see exactly what Claude called.

Connect on LinkedInSagnik Bhattacharya

It does not appear in the Shopify App Store. Shopify's help centre says you install it from Claude, which then sends you to your Shopify admin to approve the data access it asks for. No code is involved, and Shopify recommends a desktop browser for the setup. The order that works well:

Subscribe on YouTube@sagnikteaches

1. Sign in to Claude with the account you will use for store work. On a Team or Enterprise plan, an Owner has to enable connectors for the organisation first; Claude's help centre says members who cannot add one get a Request option on the connector's directory page instead.
2. Open Claude's connector settings and add Shopify from the directory.
3. Claude opens your Shopify admin. Sign in as the person whose permissions you want Claude to borrow (see the staff account section below).
4. Read the access list on the approval screen line by line. Approve read access only, unless you already have a tested write task.
5. Return to Claude and test with a question that names Shopify and your shop, such as "In Shopify, how many orders did we get last Tuesday?"

Two behaviours worth knowing before you click approve, both stated in Shopify's help centre. First, access can be raised later (you are prompted to approve the increase) but never lowered: to tighten it you uninstall the integration and install it again. Second, Claude can do no more than the person who connected it. Access is capped by that user's own permissions, so a staff login that cannot see customers or edit products will not give Claude those powers either.

Neither the directory entry nor Shopify's page names a Claude plan, and Claude's help centre says web connectors are available to all users, so even a Free account can connect. Shopify says there is no extra Shopify fee for using the connector. For reference, claude.com/pricing lists Pro at $20 a month, or $17 a month billed annually, and Team at $25 a seat billed monthly or $20 billed annually, for 2 to 150 seats. If several people will use it, the Claude Team cost breakdown shows when a shared plan beats individual ones.

Reading your store: what Claude can answer straight away

With read access, Claude can pull the following from Shopify. The tool names come from Claude's connector listing; the examples are how I would use each one.

• Orders. "List unfulfilled orders above $150 placed more than three days ago." Claude calls the list-orders tool, then sorts and summarises the result in plain English.
• Products and catalogue. "Which products have no description over 40 words?" or "Search for every product tagged outdoor with fewer than 3 photos."
• Inventory. "Which variants have fewer than 10 units across all locations, and how many sold in the last 30 days?" That second half joins two sources inside Shopify, which Claude does well.
• Customers. "How many customers placed a second order within 60 days of their first?" The listing names a list-customers tool and a get-order tool, so repeat-purchase questions are within reach.
• Analytics. The listing includes a run-analytics-query tool and says the connector pulls analytics on your store's performance, so weekly sales or sales by channel are fair questions.

Everything above is data Shopify already holds. The connector does not add intelligence to your store; it removes the clicking. A question that takes eight minutes of filters and exports in the admin takes one sentence in chat. Anything the connector cannot reach still has to come in as an exported file, which is where the gaps further down get filled.

Asking so the connector is used, not guessed

If you do not name Shopify, Claude may answer from general knowledge or from a file in the chat. Start prompts with "In Shopify" or "Using the Shopify connector", give a date range with start and end, and name the product or collection if the question is narrow. Narrow questions are also easier to verify. Shopify's help page adds that analytics questions work best short and specific, one comparison at a time with follow-ups, because a question that layers several comparisons and filters can fail. Compare these:

Weak:   How are we doing?
Better: Using the Shopify connector, give me net sales, order count and
        average order value for 1 to 31 October, and the same for
        1 to 30 September. Show the change as a percentage.

A typical illustrative reply to the better prompt: "October: net sales $41,200 from 612 orders, average order value $67.32. September: $36,450 from 575 orders, $63.39. Net sales up 13.0 percent, orders up 6.4 percent, average order value up 6.2 percent." You would then open Shopify's sales report for October and check that the $41,200 matches. If it does not match, the cause is nearly always a date boundary or whether discounts and returns are included in "net".

Changing your store: what write access adds and what stays blocked

If you approve write access, the listing's tool names show what becomes possible: creating a product, updating a product, setting the available quantity of an item at one location (with a reason such as received or damaged), adding items to a collection, creating a percentage-off discount code for the whole store, chosen products or a collection, creating or editing a collection, uploading an image, and switching up to 50 products at a time between active, draft and archived. A tool that sends a raw write query to Shopify's data interface is listed too, which is the broadest of them all.

Shopify's help centre lists what stays blocked even with write access: issuing refunds, cancelling or capturing orders, marking orders as paid, processing returns, and creating or adjusting gift cards. The connector also cannot edit existing orders or customer records, change existing discount codes, create fixed-amount discounts, or touch your plan, payments, taxes or domains. Money movement stays with a human. The list can change, so reread that page before you plan a write task around it.

Shopify states that actions connected AI tools take on your behalf are not reviewed or approved by Shopify. If Claude sets the wrong stock number or archives 40 products, that is a change in your store that you own. There is a mechanical limit too: the tool changes items one at a time, so a large batch can stop partway and has to be finished in the admin. So write access suits tasks you can check in a minute and undo:

• Drafting a new product with title, description, tags and one image, saved as draft status rather than active.
• Creating a percentage discount code like WELCOME10 (fixed-amount codes are off the list). Shopify says the connector asks you to confirm the start date and who can use the code; add an end date or a one-per-customer limit in your Shopify admin if the chat does not offer them.
• Putting 14 named products into a seasonal collection.

Tasks I would not hand over on day one: bulk price edits, theme work (the connector edits only unpublished themes; Shopify's advice is to duplicate the live theme, let the tool edit the copy and publish it yourself) and anything touching more than 25 products at once, given that batches can stall partway. For store tasks done inside the admin itself, Shopify Sidekick is the vendor's own route and has its own limits.

The four things the connector cannot see

This is the core of the title. The connector sees what Shopify holds and nothing else: your ad accounts, carrier invoices and returns paperwork are separate systems it has no tool for. So any profit answer stops at sales minus the cost per item you typed into Shopify. Those four numbers, and where each really lives:

The same logic covers payment processing fees if you ever ask for "profit": the percentage and fixed fee on each order come from your payment provider's statements, not from the product line. Add subscriptions to apps and your own time and the gap widens. Claude is not wrong when it says a product earned $12,600. It is answering the question it was able to ask.

Returns need one distinction. Shopify blocks the connector from processing returns, which is a different thing from reading them. Returns handled in Shopify show up in its sales report as sales reversals, and net sales already subtracts them, so the analytics tool may be able to report the refunded value. What Shopify does not hold is what each return cost you to handle. Test it: ask "In Shopify, what were sales reversals for September, and which products had the most units returned?" and compare the answer with Shopify's sales report. If the numbers agree, you have the refund side. If they do not, export returns yourself.

One margin question answered twice, then with the missing numbers

An illustration. Take a small shop selling printed cushions, with three products and one quarter of data. In Shopify, the owner has filled the "Cost per item" field for each variant, which is where the connector gets cost from. Asked "Using Shopify, rank my three products by gross margin for July to September", Claude returns something like this (figures are illustrative):

On this table the velvet cushion wins and the floor cushion trails. Now add the outside numbers the owner exports by hand, one CSV each from the ad platforms, the shipping app and the returns report:

Check the sum for the linen row: 7,980 minus 3,900 minus 1,470 minus 1,140 equals 1,470, and 1,470 divided by 12,600 is 11.7 percent. Velvet stays first, but linen drops from second place to last, below the floor cushion, and the conclusion flips from "linen is a solid number one by sales and a good margin" to "linen is carried by $3,900 of advertising and 38 returned units out of 420". The shop's decision is different: cut linen ad spend by a third and test, rather than order more linen.

The workflow that produced the second table is the pattern to copy: let the connector deliver the Shopify side, export the three outside files, upload them to the same chat, and ask Claude to join them by product. The step-by-step version of that margin build is in checking your margins product by product with AI.

Prompts to run in your first hour, with what good output looks like

These are read-only. Each has a "check" line, because the connector's answers are only as good as your habit of testing them.

Stock that will run out before the next order arrives

Using the Shopify connector, list every variant where available stock
across all locations is under 15 units. For each, show units sold in the
last 30 days and days of cover (stock divided by average daily sales).
Sort by days of cover, lowest first. Only show variants that sold at least
5 units in the period.

An illustrative reply: "Eleven variants qualify. Lowest cover: Velvet cushion, sage, 6 units, 41 sold, 4.4 days. Linen cushion, natural, 9 units, 58 sold, 4.7 days." Check by opening the inventory page in Shopify and filtering by the same variant. If lead time from your supplier is three weeks, the table is the reorder list. If the connector reports stock for a location you closed, tell it to exclude that location by name. If a layered request like this one fails or comes back muddled, split it: ask for the low-stock variants first, then for the last 30 days of sales on just those.

Customers who bought once and have gone quiet

In Shopify, find customers who placed exactly one order between 1 January
and 30 June and none since. Give the count, their average first-order
value, and the three products most often in those first orders. Do not
list names or email addresses.

The last sentence matters. Asking for aggregates keeps personal details out of the chat. Then ask for the three products and the first-order value, and you can brief an email without ever pasting a customer list. Check the count against Shopify's customer segments.

Product pages that are letting the shop down

Search my Shopify products for items with fewer than 3 images, a
description under 40 words, or no product type. Return a table with the
product title, which of the three problems applies, and the number of
orders in the last 90 days. Sort by orders, highest first.

This is the kind of audit that justifies connecting: a bestseller with one photo and no description is the cheapest fix in the shop. Fix the top ten by orders first; a thin page on a product nobody buys can wait.

Why orders stall

List Shopify orders that are paid but unfulfilled and older than 48
hours. Group them by shipping method, then by whether any item is out of
stock. Tell me which group is largest and what the likely cause is,
but say clearly where you are guessing.

An honest reply separates "out of stock on one line item" (a fact in the data) from "probably a courier collection problem" (a guess). Insist on that separation in the prompt, as shown, because a connector that can read the data can also produce a neat story the data does not support.

When Claude answers about the wrong store, or the wrong dates

Most connector errors are not clever. They are three plain slips. A store owner with two shops (an illustration) asks about September sales, gets $18,900, and quotes it in a supplier meeting. The figure belonged to the second shop, which was still the active connection from the day before. The connector listing includes a switch-shop tool and Shopify's help centre says you connect one store at a time, so the lesson is to name the shop in every prompt and have Claude print the store's permanent address in its first line of the answer.

The second slip is the date range. "Last month" can mean the calendar month or the last thirty days, and where a day starts depends on the time zone applied. Give start and end dates, and ask Claude to state the time zone it used. The third is definition drift: gross sales, net sales and total sales are different columns in Shopify. Gross sales is price times quantity; net sales takes off discounts and sales reversals; total sales then adds taxes, duties, shipping charges and fees. Ask Claude to state which one it used. If you want a figure to reconcile to the penny with your accounts, export the report yourself and use the connector for exploring, not for the number you file.

A fourth, rarer problem is volume. A very large question ("every order since 2019 with customer lifetime value") can run into limits on how much the connector returns in one go: Shopify's help page says the list-orders action returns up to 50 orders at a time, and Claude may summarise a slice without saying so. Ask it to state how many orders it read. If the count is exactly 50, or another round number below your real order count, it probably read one page of results and stopped. Rephrase the question as a total or count for a date range, which the connector answers from your store's analytics data rather than from a page of orders. Shopify's own order export is a related limit: it downloads up to 50 orders directly and emails anything larger, so a bigger export arrives later as an emailed file.

The staff account that limits what Claude can do

Because the connector borrows the permissions of whoever signs in, the safest move is to not sign in as the owner. Create a staff account for the purpose and give it the minimum:

• Orders: view only, if you want sales questions answered.
• Products and inventory: view only, until you have a tested write task.
• Customers: no access, if you only need totals. Your analytics and order permissions may still return counts without personal records.
• Settings, payments, apps and users: none.

Shopify's staff permission names change, so open the staff account's permission screen and confirm the exact list rather than copying mine. The principle is the thing: a limited staff account keeps Claude limited, and if the connection is ever misused, the damage is capped by that account. Shopify's guidance is to monitor your store for changes a connected tool makes, so after any write task ask Claude to list every item it changed and open each one in the admin.

When your business has several people using Claude, the shared place for the prompts above is a Project. Setting up Claude Projects as a shared team assistant shows how to keep your date conventions, store name and "say where you are guessing" rules in one place so every prompt starts from the same instructions.

Where your data goes when you connect

Connecting means store data travels into a Claude conversation, and from there it is governed by Anthropic's terms for your plan, not by Shopify's; Shopify's own help centre makes the same point. Claude's pricing page lists model training as "None by default" on Team and Enterprise, and as opt-out on the individual plans, which have a model-improvement switch in privacy settings. If your orders include customer names, addresses and phone numbers, decide what you want in a chat before you approve any scope that includes customers. The simplest policy I use for owners is aggregates in chat, personal details out of chat: ask for counts, averages and product names, and only fetch an individual's record when you are about to act on that customer.

Claude's own directory carries a plain warning to use only connectors from developers you trust. This one is made by Shopify and verified by Anthropic, but the warning is a reminder that every connector is a door. The wider tour of what a connected assistant can reach in other tools is in connecting ChatGPT or Claude to your business apps.

Filling the gaps: what to export, how often, and a monthly routine

A month-end routine that takes around 40 minutes once it is set up, using Claude for the arithmetic and your eyes for the checks:

1. Ask the connector for the Shopify side. Net sales by product, units, discounts given, stock levels, and cost per item. Save the reply as a table.
2. Export advertising spend. One CSV from each ad platform, by campaign or by product set, for the same dates. Rename campaigns so a product name appears in them; otherwise Claude has to guess which advert sold which product.
3. Export fulfilment cost. From your shipping app or carrier invoice: label cost per order. Subtract what customers paid for shipping to get the cost you absorbed.
4. Export returns. Units returned and refunded value by product, plus any return labels you paid for.
5. Upload the three files to the chat and ask for the join. "Join these by product title, calculate real profit per product and real margin, and flag any row where the ad spend could not be matched to a product."
6. Spot-check two rows by hand. Pick the best and the worst; redo the sum with a calculator. If both match, trust the rest of the table.

The unmatched-ad-spend flag in step 5 earns its place. In a typical month, a part of the spend cannot be tied to one product: brand campaigns, retargeting, a sitewide sale. Decide in advance how you split it (by share of sales, equally, or kept as an overhead line) and write that rule into the Project instructions so the answer is the same each month.

For one-off cleanups of messy export files before they are uploaded, cleaning messy data with AI covers the common problems: merged cells, dates as text, and the trailing totals row that Claude otherwise counts as a product. Another habit that pays: put cost per item into Shopify for every variant. Shopify's own profit reports simply leave out variants with no recorded cost, while an AI doing its own sums may treat a blank as zero, and a product with no cost then looks like a 100 percent margin. Ask Claude "which variants have no cost per item?" as your first connector question and fix those before you trust any margin answer.

How to tell the connector is working properly

Run this four-question test the day you connect, and again after any change in permissions:

1. "In Shopify, how many orders were placed yesterday?" Compare with the orders page, same time zone.
2. "Which three products had the most units sold last week?" Compare with Shopify's top-products report.
3. "What is the available stock of [a product you know]?" Compare with the product page, and check it adds up across locations.
4. "What can you not tell me about my store's profit?" A good answer lists advertising, fulfilment, returns and fees without being led. If Claude claims it can calculate profit, correct it and tell it which data it lacks.

The fourth question is the one most people skip, and it teaches the habit that matters. Claude will usually say the right thing when asked directly; it just does not volunteer the limits when you ask "what is my best product?". Put that caveat into a standing instruction once: "When I ask about profit or best products, state which costs are not included in your figures."

Disconnecting, and what to do if you approved too much

To remove access, take two steps. In Claude, disconnect Shopify in your connector settings. In your Shopify admin, uninstall the integration; Shopify says that revokes the tool's access, after which it can no longer read or edit store data. If you approved write access and want to go back to read only, Shopify confirms you cannot lower an approved access level: uninstall, then connect again and approve the narrower level. It takes a few minutes and costs nothing.

If you suspect Claude changed something it should not have, ask it in the same conversation which tools it ran on which items, check those items in the admin, restore from your own records, and then tighten the staff account behind the connector. The best protection is a short list in the Project of what Claude may and may not do: "Never change prices. Create products as drafts only. Ask me before touching more than 10 items."

If you would rather decide which of your store questions belong with Claude, which with Sidekick and which with a spreadsheet, that is the kind of mapping my AI implementation consultation covers.

Questions store owners ask before connecting

Does the Shopify connector cost extra?

Shopify's help centre says there is no extra Shopify fee for using the connector; your normal Shopify plan applies. On the Claude side, Claude's help centre says web connectors are available to all users, Free accounts included, so the cost is whichever Claude plan you already use. On Team, an Owner must enable connectors first.

Can Claude see my customers' personal details?

It can read customer records within the access you approve and the permissions of the account that connected it; the listing names tools that list customers and fetch orders. Shopify says it cannot edit existing customer records. If you do not want names and addresses in a chat, connect through a staff account with no customer access, or ask only for totals.

Can I connect two stores at once?

No. Shopify's help centre says the connector works with one store at a time; you can ask which store it is on and switch to another store you own, which needs a fresh authorisation. If it keeps asking to authorise the store you left, start a new conversation. Name the store in every prompt.

How do I disconnect Claude from my store?

Disconnect Shopify in Claude's connector settings, then uninstall the integration from your Shopify admin, which Shopify says revokes the tool's access to read or edit store data. You cannot lower an access level you already approved, so to go from write access back to read-only you uninstall and install again.

Will Claude change my store without asking?

Only within the access you approved: Shopify says the connector reads data but cannot change it unless you approve write access. Claude's connector settings also let you mark tools Always allow, Needs approval or Blocked. Shopify does not review what the tool changes, so start read-only and keep write tools on Needs approval.

Further reads

• How Much Does It Cost to Add AI to a Shopify Store? — What adding AI to a Shopify store costs in total.
• How to Spot Patterns in Complaints, Returns, and Defects With AI — Find the return patterns the connector cannot show.
• ChatGPT Connectors: What They Can See in Your Drive and Inbox — The same question for ChatGPT's connected apps.
• How to Sync Stock Across Shopify, Amazon and eBay Automatically — Stock accuracy if you sell on more than one channel.
• Shopify Magic vs ChatGPT for Product Descriptions — Compare Shopify's built-in writer with a chat assistant.
• Can an AI Chatbot Handle Order Tracking and Returns? — Customer-facing order and returns questions are a different job.
• How to Write Shopify Product Descriptions With AI That Convert — Rewrite the thin product pages the audit prompt finds.
• How to Use Claude With Excel and Google Sheets Safely — Safe habits for the spreadsheet exports you upload to Claude.
• Can AI Run My Shopify Store? What to Hand Over and What to Keep — A task-by-task split of what AI can do on a Shopify store, what you should keep, and a weekly review to tie it together.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Claude's connector listing for Shopify (claude.com marketplace); Shopify's help centre pages on the Shopify connector for Claude, authorising AI tool access, considerations for third-party AI tools, exporting orders, and the sales and profit reports; Claude's help centre article on connectors; claude.com/pricing. All checked 5 November 2026.