Claude Shopify Skill: Set Up Shopify's AI Toolkit in Claude

The quickest way to get a Claude Shopify skill is to install Shopify's own AI Toolkit in Claude Code. It bundles a single shopify skill, a local Dev MCP server and store-management commands, so Claude looks up current Shopify docs and API schemas, validates GraphQL and Liquid before handing code back, and can run Admin API operations on a store you've authenticated. On top of that, you can write your own skills for the store work Shopify's toolkit doesn't cover, such as product copy in your brand voice.

Here's what this guide covers:

• What Shopify actually ships for Claude, and how the plugin, skill and Dev MCP server relate
• Step-by-step setup in Claude Code and Cursor
• Connecting a real store with Shopify CLI, and why mutations are off by default
• Three custom skills worth writing for day-to-day store work
• Example prompts and the cautions that stop an agent from editing your live shop

If you're new to skills in general, start with how Claude skills work. Otherwise, let's get into it.

What the Claude Shopify skill actually is

There's no single "Claude Shopify skill" made by Anthropic. Whether you call it a Shopify AI skill or a Claude Shopify code skill, what you're after is Shopify's official AI Toolkit, which Shopify publishes for Claude Code, Codex, Antigravity CLI, Cursor, VS Code and a few other agents.

The toolkit does three things, according to Shopify's docs:

1. Docs and API schemas. The agent searches current Shopify reference material instead of relying on model memory.
2. Code validation. GraphQL queries, Liquid templates and Shopify extensions are checked against Shopify's schemas.
3. Store management. The agent prepares store tasks through Shopify CLI's authenticated store context, and you choose when they run.

It comes in three packages. Pick one:

The shopify skill itself is broad. Its source in Shopify's GitHub repo lists the Admin API, Storefront GraphQL API, ShopifyQL, Shopify Functions, Polaris extensions, POS UI, Hydrogen, Liquid, custom data (metafields and metaobjects) and App Store review. The skill tells Claude to search the docs before writing code, then validate every piece of code before returning it.

One heads-up: older versions shipped separate skills such as shopify-admin and shopify-liquid. Shopify has deprecated those. If you installed them, swap them for the single shopify skill.

How to set up the Shopify AI Toolkit in Claude Code

You need Node.js 18 or higher and a supported AI tool. That's the whole list of requirements.

Option 1: the plugin (one command)

Shopify's docs give this command for Claude Code:

claude plugin install shopify-ai-toolkit@claude-plugins-official

That's it. The plugin updates itself as Shopify adds capabilities, which is why Shopify recommends it over the other two routes.

In Cursor, type /add-plugin shopify in Cursor Chat instead.

Option 2: the Dev MCP server only

If you only want docs search and validation, add the Shopify Dev MCP server. It runs locally over stdio and needs no authentication.

In Claude Code:

claude mcp add --transport stdio shopify-dev-mcp -- npx -y @shopify/dev-mcp@latest

Then restart Claude Code.

In Cursor, go to Settings > Cursor Settings > Tools and MCP > New MCP server and add:

{
  "mcpServers": {
    "shopify-dev-mcp": {
      "command": "npx",
      "args": ["-y", "@shopify/dev-mcp@latest"]
    }
  }
}

Save and restart Cursor. On Windows, if Cursor reports connection errors, Shopify's docs suggest using cmd as the command with /k, npx, -y and the package as arguments.

Check that it works

Ask Claude something only current docs would get right. For example:

Using the Shopify docs, write a GraphQL Admin API query that returns the first 10 products with their variants and inventory quantities. Validate it before you show me.

A working setup searches the docs, writes the query, runs validation and tells you the result. If Claude answers instantly from memory with no tool calls, the server or plugin isn't loaded.

Liquid validation has two modes. By default the Dev MCP server exposes validate_theme, which checks a whole theme directory. Set the environment variable LIQUID_VALIDATION_MODE to partial if you'd rather validate individual code blocks.

Connecting Claude to a real store

Docs and validation don't touch your shop. Store management does, and it goes through Shopify CLI, not the Dev MCP server.

Two commands matter:

• shopify store auth authenticates against one store. It requires --store (the myshopify.com domain) and --scopes, a comma-separated list of Admin API scopes. It saves an online access token on your machine for later commands.
• shopify store execute runs an Admin API GraphQL query or mutation against that store, using the saved token.

The important detail: mutations are disabled by default. store execute only changes store data when you pass --allow-mutations. Shopify's docs also note that the --verbose output may include sensitive data.

So the safe pattern looks like this:

1. Authenticate with read scopes only at first, for example read_products or read_orders. Shopify lists every scope in its access scopes reference.
2. Let Claude write and run read queries freely. Reporting and audits need nothing more.
3. When you need a change, have Claude write the mutation to a file, read it yourself, and only then run it with --allow-mutations.
4. Re-run store auth with a write scope such as write_products only for the job that needs it.

Better still, do steps 1 to 3 on a development store before you point anything at your live shop.

Write your own Shopify skills for store work

Shopify's skill knows Shopify. It doesn't know your store: your tone, your theme's quirks, which collections matter. That's where your own skills come in.

A Claude Code skill is a folder with a SKILL.md file. Per Anthropic's skills documentation, personal skills live in ~/.claude/skills/<skill-name>/ and project skills in .claude/skills/<skill-name>/ inside a repo. Claude loads a skill automatically when your request matches its description, or you call it with /skill-name.

Here are three worth building.

1. Product descriptions in your brand voice

This is the skill most store owners want first. Write down the rules your best copy already follows, and Claude applies them every time.

---
name: product-copy
description: Write or rewrite Shopify product descriptions in our brand voice. Use when asked for product copy, PDP text or product page SEO.
---

Voice: warm, specific, no hype words ("amazing", "game-changing").
Structure: one-line hook, three benefit bullets, materials and care, sizing note.
Length: 80-140 words. Use the product's real materials and dimensions only.
Never invent certifications, awards or reviews.
Output the description as HTML for the descriptionHtml field, plus a
meta description under 155 characters.
See examples.md in this folder for five approved descriptions.

Put five approved descriptions in an examples.md file next to it. Examples teach voice better than adjectives do.

2. Liquid theme edits

Theme work goes wrong in predictable ways: edits to the live theme, broken sections, missing translations. A skill can encode your house rules:

• Work only in a local copy of the theme pulled with Shopify CLI, never in the theme editor
• Change one section or snippet at a time and run theme validation after each change
• Keep text in locale files, not hardcoded in templates
• Summarise every changed file before you preview anything

Pair this with Shopify's Theme Check linter and the Liquid reference, which the shopify skill already searches.

3. Admin API scripts

For bulk jobs like tagging products, fixing SKUs or exporting inventory, a skill can set the guardrails Shopify's tooling leaves to you:

• Always run a read query first and show how many records will change
• Write mutations to a .graphql file with a matching variables file
• Never pass --allow-mutations without the user's explicit go-ahead in the current conversation
• Process in small batches and log every changed ID to a file

Claude Code's disable-model-invocation: true frontmatter field is useful here. It means the skill only runs when you type its name, so Claude never decides on its own to start a bulk edit.

Not sure whether a job needs a skill or an MCP server? The short answer: skills hold instructions, MCP servers hold connections. Our skills vs MCP comparison goes deeper.

Example prompts for Claude and Shopify

Once the toolkit and your skills are in place, these prompts cover most day-to-day work:

The last prompt comes straight from Shopify's docs, which add that the toolkit doesn't remove the need to review and test the result.

Storefront MCP: for shopping agents, not store admin

Shopify runs a second family of MCP servers aimed at buyers rather than developers. Storefront MCP lets an AI agent search one merchant's catalog through an endpoint on the store's own domain, with tools such as search_catalog, lookup_catalog and get_product. Each request carries a profile URL that identifies the agent. For discovery across many stores, Shopify points to its Global Catalog MCP instead.

This matters if you're building a shopping assistant for your storefront. It's not the tool for editing products or themes; that's the AI Toolkit's job. If you want more MCP servers beyond Shopify, see our list of the best MCP servers.

Cautions before you give Claude your store

Agents are fast, and fast mistakes on a live store cost money. A few rules:

• Use a development store first. Test every skill and script there before it touches real products or orders.
• Request the narrowest scopes. Read scopes for audits and reports; add a write scope only for the job that needs it.
• Keep mutations behind a human. Shopify's default of mutations off is the right one. Don't script around it.
• Never edit the live theme directly. Duplicate it, work on the copy, preview, then publish.
• Check what gets logged. Per the toolkit's README, telemetry is on by default and can include search queries, validated code and, on some surfaces, your latest message. You can opt out by creating an empty file at ~/.config/shopify-ai-toolkit/opt-out, or by setting OPT_OUT_INSTRUMENTATION=true.
• Review generated copy for facts. A brand-voice skill makes copy sound right. It can't tell whether a fabric is really organic.

Key takeaways

• Shopify's official AI Toolkit is the "Claude Shopify skill" most people mean. The plugin is the easiest install.
• The Dev MCP server alone gives you docs search and GraphQL, Liquid and extension validation, with no store access.
• Store changes go through shopify store auth and shopify store execute, and mutations stay off unless you pass --allow-mutations.
• Your own skills fill the gap: brand voice, theme rules and safe bulk-edit procedures.

All of this assumes you're comfortable in a terminal with Node installed. If the setup is where you keep stalling, Taku mirrors an AI app or workflow someone already got working into a desktop workspace, where you can run it, edit it and save it as your own. Taku is in Beta, and the Mac app is available now.

FAQ

Is there an official Claude Shopify skill?

Yes, but it comes from Shopify, not Anthropic. Shopify's AI Toolkit includes a shopify skill that covers its APIs, Liquid, Hydrogen, Functions and Shopify CLI, and it installs into Claude Code as a plugin.

What's the difference between the Shopify skill and the Shopify Dev MCP server?

The skill is a set of instructions plus helper scripts that tell Claude how to search and validate. The Dev MCP server exposes docs search and validation as MCP tools any MCP client can call. The plugin installs both.

Does the Shopify Dev MCP server need an API key?

No. It runs locally and doesn't require authentication. Only store management through Shopify CLI needs you to log in to a store.

Can Claude edit my Shopify store directly?

Only if you let it. Claude can run Admin API operations through shopify store execute after you authenticate with shopify store auth. Mutations are blocked unless you add --allow-mutations, and the scopes you grant limit what's possible.

Does the Shopify AI Toolkit work in Cursor?

Yes. In Cursor Chat, run /add-plugin shopify, or add the Dev MCP server under Tools and MCP in Cursor's settings.

Can I use a Shopify AI skill for product descriptions?

Shopify's skill is built for development work, so write your own. A short SKILL.md with your voice rules plus a file of approved examples gives you consistent copy across the catalog.