How to Connect Shopify to Claude the Read-Only Way

You can connect your Shopify store to Claude in two ways. The fastest is the official Shopify connector in Claude's directory, which uses a sign-in and needs no token. The tighter option is a read-only Admin API token, which lets you control exactly what Claude can see. Either way, Claude reads your live orders, products, inventory, and customers, and answers questions about them in plain English. You can ask about yesterday's sales or your slowest products without exporting a file.

One thing is worth knowing before you start. Shopify stores your revenue, but not your landed cost of goods (COGS). So Claude can rank products by revenue on its own, but it cannot tell you true profit unless you give it your costs. The rest of this guide covers both setup paths and how to confirm the connection is live. It also covers what Claude can and cannot answer, and how to keep access read-only so a misread question never changes a price or a stock level.

Two Ways to Connect Shopify to Claude

Both paths end in the same place, with Claude reading your live store. They differ in how much control you have over access and how much setup they take. The table below is the short version, and the steps follow underneath.

	One-click connector	Read-only API token
Setup	Sign in and approve access	Create a custom app, copy a token
Access control	An access level you approve	The exact scopes you choose
Credential to manage	None, it uses OAuth	A secret token you store safely
Best for	Trying it, a single-operator store	Shared stores, strict least privilege

Start with the one-click connector if you just want to try it on your own store. Use the API token when more than one person has access to the store, or when you want to prove that Claude can only ever read your data.

Option 1: The one-click Shopify connector

This is the default for most store owners, and it takes about a minute.

1. In Claude, open Settings, then Connectors.
2. Find Shopify in the directory, or go to claude.ai/directory/connectors/shopify.
3. Sign in to your Shopify account and approve access for one store.
4. Start a chat and ask a question about your store.

The connector uses OAuth, so there is no token to copy or paste. During step three Claude shows you an approval screen that lists the access it wants. Choose the read-only level unless you have a specific reason to let it write. Write access would let a misread prompt change prices, stock, or discount codes. The connection applies to one store at a time, so if you run several stores you authorize each one separately. You can remove a store's access later from the same Connectors screen.

Option 2: A read-only Admin API token

Use this when you want tighter control than the directory connector's access level gives you. It is the least-privilege route, because you grant only the scopes you choose and nothing else.

1. In your Shopify admin, go to Settings, then Apps and sales channels, then Develop apps, and create an app.
2. Under Admin API access scopes, grant only the read scopes you need, and leave every write scope off.
3. Install the app and copy the Admin API access token.
4. Add the Shopify Admin connector in Claude and supply your store domain and that token.

For store analysis, these four read scopes cover almost everything you will ask about.

Scope	Lets Claude read
read_orders	Orders, line items, and discounts
read_products	Products and their variants
read_inventory	Stock on hand by location
read_customers	Customer records and order history

Treat the access token like a password, because anyone who has it can read your store data. Store it in a password manager, do not paste it into a shared document, and rotate it if it ever leaks. If more than one person has access to the store, this route is safer than the directory toggle. A token with only read scopes cannot change anything, no matter what you or the model asks it to do. For the wider habit of giving any AI read-only, least-privilege access to your store data, see how to safely connect AI tools to your data.

Check that Claude is reading live data

Before you trust any answer, run one question you already know the answer to. Ask Claude for yesterday's order count, or units on hand for a single SKU, then check it against your Shopify admin.

CopyUsing my connected Shopify store, how many orders were placed yesterday and what were the total sales? Show the figures and the exact date range you used.

If the numbers match your admin, the connection is live and reading real data instead of a stale guess. If they do not match, the cause is usually a time zone difference in the date range. It can also be a connector that defaulted to a sample rather than your live store. Ask Claude which store and date range it used, and re-check.

What Claude Can Analyze in Your Shopify Data

Once the store is connected, Claude handles the questions that only need Shopify's own data well: product performance, inventory movement, discount usage, and repeat-purchase patterns. Two examples show the range.

The numbers below are illustrative. They are sized for a mid-size apparel store doing about $120,000 a month at a $70 average order value, and are not any one brand's real figures.

The first is inventory, which Claude can answer straight from Shopify.

CopyUsing my connected Shopify store, list the products with the lowest sell-through over the last 60 days. Show units sold, units on hand, and sell-through as a percent, and include only products stocked for the full period. Return a table sorted by sell-through, lowest first.

A result might look like this.

Product	Units sold (60d)	Units on hand	Sell-through
Linen shirt, navy	18	220	8%
Wool scarf	12	140	8%
Canvas tote	40	60	40%

A sell-through near 8% points to cash sitting in stock that is not moving, which is your reorder-and-markdown list for the next cycle. Read it next to what a healthy sell-through rate looks like for your category, and see the inventory analytics guide for how one query fits a wider stock review.

The second is profit, but only if you hand Claude your costs, which is the subject of the next section.

Give Claude the Context It Is Missing

Claude only knows what the source gives it, and your profit numbers are the clearest gap. Shopify stores what you charged, not what the order cost you to fulfil. So a bare question like which products are most profitable comes back ranked by revenue, which is not the same thing. The fix is to supply your margin or your per-product costs in the prompt.

CopyUsing my connected Shopify store, list my 10 products with the highest gross profit over the last 90 days, not the highest revenue. My average gross margin is 45%; use it unless I paste per-product costs. Show units sold, revenue, estimated COGS, and gross profit as a table.

With the margin supplied, the ranking changes, and a lower-revenue item with a stronger margin can move up the list. See gross margin for the costs to include before you trust any profit ranking.

What Claude Cannot See

Beyond cost, two more limits are worth knowing before you lean on an answer.

The first is other sources. Claude connected to Shopify alone cannot work out blended customer acquisition cost (CAC), because the ad spend lives in Meta and Google. If you add those connectors and ask for CAC, a different problem appears. Say Meta claims 260 new customers and Google claims 160, which sums to 420, while Shopify shows 300 distinct first-time customers in the same month. On $18,000 of combined spend, the platform math gives a CAC near $43, but against 300 actual new customers the CAC is $60. Claude will take the platforms at their word unless you tell it to count distinct new customers in Shopify and ignore platform-claimed conversions. See blended CAC for why the two numbers diverge.

The second is depth of history. Pulling several years of orders live is slow and can hit the connector's limits, and a long date range uses a lot of the model's context budget. For a one-off question over a wide range, exporting a CSV and uploading it is often cleaner than a live pull.

We did not join those sources by hand. Our question was whether blended CAC last month was really $43 or something higher once we counted distinct new Shopify customers. Daymark connects Shopify, Meta, and Google into one place and does the cross-source math the same way every time. That way the $60 figure is the one you set a budget against, instead of the flattering $43 that over-credits your ads. See more on proving what your ad spend actually returns.

When the Numbers Look Off

A connected model will still produce a confident figure even when the data behind it is thin, so a few checks keep you honest.

• Ask it to show the fields and rows behind each number, so you can spot a wrong join or a filled gap.
• Give it the business rule it cannot infer, such as what counts as a new customer or your return window, before you ask the question.
• Confirm any figure you are about to spend or reorder against directly in Shopify first.
• If a profit or CAC number looks too good, check whether costs and ad spend were actually included, because a missing cost line is the usual cause.

Claude or ChatGPT for Shopify

Both work the same way. Shopify ships official connectors for Claude and for ChatGPT, and both read your live store. The prompts you use are nearly identical, so pick the one you already pay for rather than switching tools for this. If you are setting up ChatGPT instead, the ChatGPT Shopify guide covers that path, and the same prompt library works in either one.

What to Do This Week

Connect your store with read-only access. Then ask Claude a few questions you already care about, such as your slowest sellers, your discount codes, and your repeat-purchase rate. Before you ask anything about profit, give it your cost, shipping, and fee numbers. Before you ask about CAC, add your ad connectors and tell it to count distinct new customers in Shopify.

If you would rather not wire up and maintain each source yourself, or re-explain your costs every time you ask, you can connect your data once and let the cross-source margin and CAC math be handled for you. Daymark brings Shopify, Meta, Google, and GA4 into one workspace and answers in plain English, with a 14-day free trial and no card required. For where a connected store fits a weekly review, see tracing why your numbers changed.

Frequently Asked Questions

Can Claude connect to my live Shopify store?

Yes. Shopify ships an official connector in Claude's directory. You open Settings, then Connectors, sign in to your store, and approve an access level, after which Claude reads your live orders, products, inventory, and customers. For tighter control you can instead create a custom app with a read-only Admin API token scoped to the data you want. No CSV export is needed to get started.

What Shopify API scopes should I give Claude?

Grant only read scopes for the data you want Claude to see: read_orders, read_products, read_inventory, and read_customers. Leave every write scope off unless you have a clear reason to let the model change prices, stock, or discounts. Least privilege means a misread prompt cannot alter your store. The one-click connector uses OAuth instead of a token, but choose its read-only access level for the same reason.

Is the one-click connector or an API token better?

The one-click connector is faster and fine for a single-operator store, because it signs in with OAuth and needs no token to manage. A read-only Admin API token is better when more than one person has store access or you want strict least privilege, since it grants only the exact scopes you choose and cannot change anything. Both keep access read-only when you set them up that way.

Why does Claude get my Shopify profit numbers wrong?

Because Shopify stores revenue, not your landed cost of goods. A model reading raw orders ranks products by revenue and labels it profit, so your top seller by units can look like your top earner when it is not. Supply your COGS, shipping, and fee assumptions in the prompt, and ask Claude to show which fields it used, so you can check the result before acting on it.

Is it safe to connect Shopify to an AI assistant?

It is when you keep access read-only. Grant read scopes only, or approve the read-only access level on the one-click connector, so the model can read data but cannot change a price, a discount, or stock. Verify a known number after connecting to confirm it reads live data, and confirm any figure you plan to spend or reorder against directly in Shopify first.