Short answer: if you have no budget and a simple product, use Claude with a good theme and don’t hire anyone, including me. If you’re turning a consistent profit and the store is the thing holding you back, hire a designer. And if you’ve already built it with Claude and it looks great but isn’t selling, you don’t need a rebuild, you need someone to find the leak.

That’s the whole post. The rest is the maths behind it, because ‘it depends’ is the answer most designers give and I’d rather you had the actual numbers.

Route one and two

What does it actually cost to build a Shopify store with Claude?

On paper, nearly nothing. A Claude subscription, a Shopify plan you were paying for anyway, and a theme. If you buy a premium theme from the Shopify store you’re looking at a one-off £200 to £400. If you go with Dawn, which is free and perfectly good, it’s £0.

The real cost is time, and it splits into two very different amounts depending on which way you go.

Route A: theme plus Claude as your assistant. You buy a theme, and you use Claude to write product descriptions, plan your pages, generate the FAQ, tidy your About page and tell you what’s missing. This is fast. A weekend, maybe two if your product photography isn’t ready. I genuinely recommend this to start-up brands and I’ve said so since May. Nothing in this post changes that.

Route B: Claude builds the actual theme. This is the version Instagram is selling you, where Claude designs the site and pushes the code straight into your store. It works now. It didn’t in May, and it does in September, because you can connect Claude directly to your Shopify theme through GitHub or a code editor.

But it is not a weekend. Getting that connection set up was, in my own words at the time, bloody difficult, and I do this for a living. Then there’s the design itself. Claude’s first draft will be beautiful and it will be wrong in ways you can’t see yet: text that doesn’t scale on mobile, sections with too much air, a product page with no reviews, no trust signals and one photo. You’ll spend evenings training it like an intern on their first day. Realistically that’s 30 to 50 hours for someone reasonably technical, and ‘reasonably technical’ means you don’t flinch at the words ‘command line’.

Route three

What does it cost to hire a Shopify designer?

I can only quote my own prices, because the market is anywhere from £500 on a freelancer site to £15,000 at an agency and those two things are not the same product.

My Shopify Face Lift starts at £2,000 and takes 4 to 6 weeks. That’s a fully custom design and build, up to two custom product templates, plugins chosen for speed rather than habit, and the home, category, About, FAQ and blog pages done. Fees split 50% on booking and 50% on completion.

Here’s the part the price list doesn’t say. I use Claude on every one of those builds now. It’s why the things I used to gently talk clients out of, the subscription flow, the custom product labels, the animation that carries the whole brand, are on the table at £2,000 rather than at five figures. You are not choosing between AI and a human. You’re choosing whether the human driving the AI is you, at midnight, or someone who already knows where the potholes are.

The comparison

Claude vs a Shopify designer: the comparison table

The middle column is the interesting one. It’s the cheapest route to a truly custom store and the most expensive route in evenings. Whether that’s a bargain depends entirely on what your evenings are for.

The blind spot

What Claude won’t tell you about your store

Claude is a brilliant designer and a terrible shopkeeper.

When I asked it to design a homepage in May, it produced something I’d have been proud to make, in minutes. What it didn’t include, because I hadn’t asked, was any social proof, any trust signals near the add to cart button, or any of the photo types that actually move people from looking to buying. Perfect UX, zero sales instinct. And a founder building their first store would have no way of knowing what was missing, because the page looks finished.

That’s the whole gap. Baymard Institute’s research puts cart abandonment at around 70%, and most of that isn’t price. It’s friction: a checkout with a surprise, a product page that doesn’t answer the one question the buyer had, a mobile layout that hides the button. Claude will build you a store with all of that in it and tell you it’s done.

The decision

Which route should you choose?

Three founders, three answers. And a fourth, because it’s the email I get most.

You’re pre-launch or under £2,000 a month in sales. Theme plus Claude. Spend the money on product photography and a couple of months of ads instead. Do not hire me. If someone quotes you £2,000 for a store at this stage, ask them how it’ll pay for itself, and listen carefully to the answer.

You’re profitable, the DIY site is embarrassing you, and you’ve got roughly 40 free hours and a taste for it. Route B is a real option now, and I’d rather you did it well than badly. Read the connection guide, use GitHub so you can roll back, and get a product page framework before you let Claude design a single section.

You’re profitable, the site is the bottleneck, and your evenings are already spoken for by formulations, packaging suppliers and the wholesale enquiry that’s been sitting there four days. Hire a designer. That’s the Face Lift, and if you want to know whether it’s the right package, or whether any package is, that’s exactly what the discovery call is for. No jargon, no hard sell, and I’ll tell you if you’re better off with a theme.

You already built it with Claude. It looks lovely. It isn’t selling. Don’t rebuild. Something specific is leaking and it’s usually three or four settings, not the design. That’s what the Website Botox is: an audit of what you’ve got and small, targeted fixes where they’re needed, from £300, done in 1 to 2 weeks. It’s the cheapest thing on my menu and it’s the one I’d send my own sister to.

Frequently asked questions

If you’d rather have someone else do the arguing with Claude, book a discovery call. Free, no pressure, and I’ll tell you honestly which of the three routes above is yours.

There are five ways to connect Claude to your Shopify store, and they have almost nothing to do with each other. Here is the short answer before anything else.

Both get called “connecting Claude to Shopify” and the gap between them is roughly the gap between a bicycle and a forklift.

I have used four of the five methods on real client stores. Below is what each one actually does, what it costs you in setup time, and which one you should pick, including the fairly large group of people whose correct answer is “none of them, go and do something else”.

I am a Shopify designer, so I have an obvious interest in telling you this is all too hard for you and you should hire me. I am going to try not to do that. Two of these methods genuinely take under five minutes and most brand owners should turn them on today.

The short version

Which method should I use?

If you just want the answer and not the reasoning:

Some context

Why are there five methods and not one?

Shopify and Anthropic have been building towards the same thing from opposite ends, and they met in the middle in a slightly untidy way.

Shopify wanted its merchants to have an assistant, so it built Sidekick into the admin, which at the time of writing runs on Claude under the bonnet. Shopify also wanted developers to stop shipping broken Liquid, so it published a documentation server that any AI tool can read from. Anthropic, meanwhile, wanted Claude to talk to the software people already use, so it built a connectors directory with Shopify in it.

And then a whole ecosystem of third-party apps appeared to fill the gaps, because that is what always happens.

So you end up with five doors into the same building. They are all labelled “Claude” and “Shopify” and none of them tell you which room you will end up in. Below, the rooms.

Method 01

How do I connect Claude to my Shopify store directly?

Best for: almost every brand owner

This is the official Shopify integration inside Claude itself. You authorise it once, and after that you can open a normal Claude conversation and ask things like “which products had the worst sell-through in the last 60 days” and get an answer from your live store data rather than a polite guess.

Step by step

1. Open Claude, either the web app or the desktop app. Not the phone one if you can help it, because the authorisation flow is nicer on a big screen.
2. Go to Settings, then Connectors.
3. Find Shopify and click to add it.
4. You will be bounced to Shopify to log in and approve access. Read this screen. Genuinely read it. It is the only point in the process where you get to decide what Claude is allowed to do.
5. Choose read-only for now. I will explain why in a moment.
6. Pick the store you want to connect. It is one store at a time, so if you run three you will be switching.

That is it. It is free to turn on. You are not paying Shopify anything extra for it, though what data you can reach still depends on your Shopify plan.

What it can actually do

Reading is where it shines. Orders, products, inventory levels, customers, discount codes, analytics. You can ask it to summarise a month, compare two periods, find the customers who have ordered three times and never used a discount, or work out which collection is quietly carrying the whole store.

With write access granted it can also update products, adjust inventory quantities and create discount codes. This is useful and it is also the bit where you want to be a grown-up about permissions.

What it cannot do

It only exists inside a conversation. Nothing is watching your store. It will not tell you when stock runs low, it will not flag a weird order at 3am, and it will not run every Monday morning unless you sit down and ask it to. If you close the tab, nothing is happening. People get this wrong constantly and then wonder why their “AI store monitoring” never alerted them to anything.

It also cannot touch your theme. If you ask it to make your product page look better it will give you advice, which is not the same as a product page that looks better.

Method 02

What is Shopify Sidekick and do I need to connect it?

Best for: small jobs you are already in the admin for

You do not need to connect anything. Sidekick is Shopify’s own assistant, already sitting in your admin, available across plans. It is built on Claude, which is why it keeps turning up in articles about connecting Claude to Shopify despite not being a connection at all.

It is good at the things it is good at: pulling up a report you cannot remember the name of, drafting product descriptions, making a discount, adjusting a few theme settings, answering “how do I do X in Shopify” without you opening a support doc.

It is less good the moment a task crosses more than one part of the admin, or needs it to hold a lot of context about your specific store. Its write capabilities are real but conservative, which for something with a direct line into a live shop is the correct design decision even when it is annoying.

Method 03

What is the Shopify AI Toolkit?

Best for: anyone writing Liquid, GraphQL or app code

This is the one most merchants have never heard of and most developers should be using.

Shopify publishes an open-source toolkit that gives an AI tool direct access to Shopify’s real documentation, its GraphQL schemas, and a validator that checks generated code against them. It grew out of the Dev MCP server Shopify shipped at the end of 2025, and it works with Claude Code, Cursor, VS Code and a few others.

The problem it solves is specific and enormous. Without it, an AI writing Liquid will occasionally invent a filter that does not exist. It writes something that looks completely plausible, ships, and then silently renders nothing on mobile. With the toolkit connected, the model reads Shopify’s actual schema instead of remembering it.

Setting it up in Claude Code

The plugin route is the one to use, because it updates itself when Shopify ships new capabilities:

/plugin marketplace add Shopify/shopify-ai-toolkit
/plugin install shopify-plugin@shopify-ai-toolkit

Documentation lookups and code validation work with no store connection at all. Nothing to authorise, no token, no risk. If you want it to run operations against a live store you will authenticate through the Shopify CLI, which sends you through a normal approval in your admin.

While you are there, Shopify also publishes a separate Liquid plugin for Claude Code that teaches it theme conventions, CSS and JS standards, and accessibility patterns for ecommerce components:

/plugin marketplace add Shopify/liquid-skills
/plugin install liquid-skills@liquid-skills

The accessibility one is worth having even if you think you know what you are doing. I thought I knew what I was doing.

One thing it is not: a way to run your shop. It is a developer tool. “Put everything in the sale collection up by 10%” is not what this is for.

Method 04

Can Claude actually edit my Shopify theme?

Best for: building and editing themes properly

Yes, but not through any of the methods above. To let Claude properly edit a theme, it needs to work on the theme files themselves. That means Claude Code, the Shopify CLI, and ideally GitHub sitting in the middle. This is the one that changed how I work.

Everything above is Claude talking about your store. This is Claude working inside it.

What I was doing before, and why it was silly

For months my process was: describe a section to Claude, get Liquid back in the chat window, copy it, open the Shopify code editor, paste it into a new section file, save, preview, spot the problem, go back to the chat, explain the problem, get a new version, copy, paste, save, preview.

It worked. I built a genuinely nice animated homepage that way and kept full control over the details. But it felt like training an intern on their first day, and it took roughly as long as building the thing from scratch would have.

The reason it was slow is obvious in hindsight. Claude could not see the theme. Every request started from nothing. It did not know what my snippets were called, what my design tokens were, what the section above did, or which of my sixteen colour variables I had actually used.

The setup that fixed it

Put the whole theme on your machine, in a git repository, with Claude Code able to read all of it. I found this route via a YouTube video by Brendan Gillen. Here is the plain-English version, because this is not that kind of blog and the exact commands move around every few months.

1. Install the Shopify CLI. One command, npm install -g @shopify/cli, and it is a prerequisite for everything below.
2. Pull your live theme down with shopify theme pull. You now have the whole thing, sections, snippets, templates, config and locales, as folders you can actually see.
3. Put it in a private git repository and commit it before you let anything near it. This is not optional. It is the step that means every mistake from here is a thirty-second revert instead of a bad afternoon.
4. Connect that repository to Shopify. Shopify’s GitHub integration lets you attach a branch to a theme, and it syncs. This is what turns “Claude wrote some code” into “the change is in my store”.
5. Open Claude Code in the theme folder so it can read the whole structure.
6. Install the toolkit and Liquid plugins from method 3, so it is writing against real schemas.
7. Write a CLAUDE.md file in the root of the repo. More on this below, because it is the single highest-value thing in the setup.

Then you type things like “read sections/main-product.liquid, create a new section that does the same thing but with the gallery on the right and sticky on scroll, keep all the existing schema settings” and it does it, in the actual files, and you preview it locally with shopify theme dev.

Was it hard?

It was bloody difficult, and I say that as someone comfortable with code. The first afternoon was mostly error messages. But it is a one-off cost. Once it is set up, the whole thing runs in plain language.

GitHub also gives you something the theme editor never has: every version of the site, forever, with the ability to go back to any of them. I have stopped duplicating themes and naming them things like LIVE - do not touch - v4 FINAL. That alone was worth the afternoon.

Claude is only as good as what you feed it

This is the thing I would tell you if we only had two minutes.

I spent a while asking Claude to design things and being underwhelmed. Then I stopped doing that. Instead of asking it to design a product page, I gave it the layout of the perfect product page and asked it to build that using the existing theme’s design language. The difference between those two outputs is not subtle. The first is a competent template. The second was the first time I have been able to build a Shopify page exactly the way I wanted it, down to the last pixel.

CLAUDE.md is where you make that permanent. It is a plain text file in the root of your repo that Claude reads automatically, and it should be short, a hundred lines or so rather than five hundred. Mine covers:

• The stack. Which base theme, how CSS is organised, which conventions I am using.
• Naming rules. How sections are named, which metafield namespaces exist, whether strings need to go through the translation files.
• The commands. How to run the dev server, how to lint, how to push.
• A file map. What is in each folder and what the prefixes mean.
• Guardrails. Never push to the live theme. Never delete a theme. Always specify metafield types.

Free template

The Perfect Homepage Builder

Speaking of giving Claude a good layout to work from. This is the homepage structure I use on client builds, section by section, free to take.

Get the template →

The rules I now work by

• Scope by section, never by epic. “Convert this theme into a custom build” produces sludge. “Read this file, make this specific change, preserve the schema” produces a diff I can review in fifteen seconds.
• Push to an unpublished theme. shopify theme push --unpublished creates a duplicate. Look at it. Then publish.
• Run shopify theme check and require zero errors before anything is considered done. This one rule prevents a whole category of broken storefront.
• Read the diff. Fifteen seconds of looking at what changed catches nearly everything. Not reading it is how people end up with stories about AI destroying their site.

What this made possible

The reason I care about all of this is not speed. It is that I can now build a store exactly the way I want it and then hand it over to a client to run themselves without any trouble.

That combination did not exist before. Previously I could either custom-code something and make it unusable for anyone without a computer science degree, or work inside a template and accept its limits. Now the theme editor still works for the client, every section is properly configurable, and the design is not compromised. If you want the design side of that argument, I have written separately about how to make your Shopify site look expensive.

Method 05

Should I use a third-party MCP app?

Best for: people who will never open a terminal

There are apps in the Shopify App Store that act as a bridge between your store and Claude. You install the app, approve the permissions, connect it in Claude, and you get a set of tools for managing products, orders, customers and inventory conversationally. There are also open-source community servers that do the same thing if you are happy configuring them yourself.

The pitch is real: more tools than the official connector, and no code. Four things to check before you install one:

• What permissions is it asking for, and does it need all of them? An app that wants write access to customers in order to help you edit product titles is asking for more than the job requires.
• Where does your store data go? You are routing your orders and customers through a third party. Read the privacy policy. I know. Read it anyway.
• Is there a confirmation step before writes? There is no dry run on a live Shopify store. Every change is real the moment it fires.
• Who maintains it? Shopify’s API changes constantly. A server that has not been updated in eight months is a liability, not a tool.

If the official connector covers what you need, and for most brand owners it does, start there instead. Fewer parties, fewer permissions, fewer things to go wrong.

The limits

What can Claude still not do with my Shopify store?

Worth saying plainly, because a lot of writing on this subject is quietly optimistic.

• It cannot watch anything. No connector here monitors your store, fires on low stock, or emails you about a fraud flag. Every one of these methods requires you to start a conversation. If you want automation, that is Shopify Flow’s job.
• It cannot fix a business problem that is not a website problem. If the product is not right, or nobody knows you exist, a better product page changes nothing. I would rather say that now than take your money for a rebuild.
• It cannot make design decisions for you. It will produce something competent from a vague brief and something excellent from a precise one. The judgment is still yours, which is either reassuring or annoying depending on why you are reading this.
• It does not know your store’s history. It does not know that the previous developer left a custom snippet half your collection pages depend on, or that the founder hates the colour green. That is what CLAUDE.md is for.
• It will do exactly what you asked. Including when what you asked and what you meant were different. That is the entire risk profile of write access in one sentence.

Decide

So which one should you actually use?

The awkward question

Did I just talk myself out of a job?

Theoretically yes, if my clients decide to spend their evenings learning Claude Code, the command line and GitHub. It is all learnable. Nothing in method 4 requires a degree.

But most of my clients are busy with product formulations, packaging suppliers, sales tax, the wholesale enquiry that has been sitting unanswered for four days, and a customer who wants to know whether the candle is vegan. They do not want to learn git. Nor should they have to.

And I will repeat the advice I gave before all of this: a start-up brand with no budget for a designer is still better off buying a good template and customising it. That has not changed and I do not expect it to.

Answers

Frequently asked questions

Where to go next

If you want the unglamorous version of all this, I wrote up my first attempt at building a Shopify store with Claude before I found the setup above, including the bits that did not work and the point where I nearly gave up. It is a useful counterweight to the tidy version you have just read: Claude Shopify: can AI really build a good website?