Automation

How to Run Your Entire Shopify Store From Claude Code: A Non-Technical Merchant's Setup Guide

April 8, 2026

If you run a Shopify store and you've been watching the AI conversation from the sidelines wondering when it actually becomes useful for you — not for developers, not for "future you who learns to code," but for the actual operator updating products and writing descriptions on a Tuesday afternoon — that moment is now.

In the last six months, three things became true at the same time:

Anthropic's Claude Code (the terminal tool that lets Claude actually do things on your computer instead of just chatting about them) became stable enough for non-developers to use safely.
Shopify shipped an official Dev MCP server that gives Claude live access to Shopify's documentation, GraphQL schemas, and API best practices.
The MCP ecosystem exploded — there are now community servers for managing your Shopify store, generating images with Google's Nano Banana, pulling Google Search Console data, posting to Klaviyo, and a hundred other things, all speaking the same protocol.

Plug them together and you get a setup where you can type "find the 20 products with the lowest CTR in Search Console, rewrite their meta descriptions, generate new lifestyle hero images for the top 5, and stage everything in a draft so I can review before publishing" — and walk away while it runs.

This guide takes you from zero installed software to that exact workflow. It's written for a Shopify merchant who is comfortable copy-pasting commands but doesn't write code. Total setup time: about 60–90 minutes the first time. After that, every new automation takes minutes.

What You'll Have at the End

By the end of this guide, you'll have Claude Code running on your computer with four superpowers connected:

Shopify Dev MCP — official Shopify docs and API knowledge built in, so Claude never invents fake fields or outdated syntax
A Shopify store-management MCP — actual control over products, orders, customers, collections, and images on your live store
Nano Banana MCP — Google Gemini-powered image generation and editing (the same model behind some of the best AI product photography you've seen this year)
Google Search Console MCP — live SEO data Claude can read and act on without you exporting CSVs

That's the full loop: see what's broken, fix it, generate the assets, push to Shopify, verify in Search Console. All from one chat window.

Why This Matters Right Now (April 2026 Context)

A few things to know before you start, because the landscape has shifted in the last quarter and most tutorials online are out of date:

As of January 1, 2026, new Shopify apps are created in the Dev Dashboard and use OAuth client credentials (a client ID and secret) instead of the old static shpat_ access tokens. If you've followed an older tutorial and it tells you to copy a shpat_xxxxx token, it'll only work for legacy custom apps. New setups use clientId/clientSecret pairs, and the MCP server exchanges those for short-lived tokens automatically.
The default Shopify API version most MCPs target is 2026-01. Any tutorial referencing 2024-04 or earlier is outdated.
Nano Banana 2 (Gemini 3.1 Flash Image) is now the default for most image MCPs — it gives you Pro-level quality at Flash-tier speed and runs roughly $0.04–$0.05 per image at standard resolution, scaling up to about $0.15 per image at 4K.
There are two Shopify MCPs and they do different things. Shopify Dev MCP is a developer tool that runs locally and provides access to Shopify documentation and API schemas. It does not control your store. For actual store actions (creating products, updating prices, managing collections) you need a separate community MCP. We'll install both.

Got it? Good. Let's set up.

Step 1: Install the Foundation (Node.js and Claude Code)

You'll need Node.js 18 or higher for almost everything in this stack. If you've never installed it:

Go to nodejs.org and download the LTS version for your operating system.
Run the installer with default settings.
Open Terminal (Mac) or Command Prompt / PowerShell (Windows) and type node --version. You should see something like v20.x.x or higher.

Next, install Claude Code:

npm install -g @anthropic-ai/claude-code

Then run it once to log in:

claude

It'll open a browser, ask you to log in to your Anthropic account, and you'll be back at the terminal with a working Claude Code session. Type /exit to leave it for now.

Heads up: Claude Code on Windows can be finicky with MCP servers — many community servers assume Mac/Linux paths. If you're on Windows and this is your first time, the smoothest path is to install WSL2 (Windows Subsystem for Linux) and run everything from inside Ubuntu. It takes 15 extra minutes but saves hours of debugging.

Step 2: Install the Shopify CLI

The Shopify CLI is the official command-line tool for working with themes, apps, and store data. Claude Code will use it under the hood.

npm install -g @shopify/cli@latest

Verify with shopify version — you should see a version number. If you want to log in to your store now (optional but useful), run shopify auth login and follow the browser prompt.

Step 3: Add the Official Shopify Dev MCP

This is the one that gives Claude access to Shopify's documentation, GraphQL schemas, and code validation. It does not need authentication and runs entirely on your machine.

claude mcp add --scope user shopify-dev-mcp -- npx -y @shopify/dev-mcp@latest

That's it. Restart Claude Code (/exit then claude again), and you can immediately ask things like:

"Show me the GraphQL mutation to update a product's metafield in API version 2026-01"
"What scopes does my app need to read draft orders?"

The Dev MCP includes a tool called learn_shopify_api that must be called before any other Shopify tools in a given conversation — it generates a conversationId that the other tools need. Claude knows this and will call it automatically. You don't need to think about it.

Step 4: Add a Shopify Store-Management MCP

This is the MCP that actually does things in your store — creates products, updates inventory, processes orders, manages collections. There are two main paths.

Option A: Community MCP (free, more setup)

The most popular open-source option is shopify-mcp by GeLi2001. It's a Node package, runs locally, and supports both legacy access tokens and the new OAuth client credentials.

For new Shopify apps (post-January 2026):

From your Shopify admin, go to Settings > Apps and sales channels > Develop apps (or the new Dev Dashboard if your store has it enabled).
Create a custom app, configure the API scopes you want Claude to have (start small — read_products, write_products, read_orders is a sensible starting trio), and copy the Client ID and Client Secret.
Add the MCP to Claude Code:

claude mcp add shopify -- npx shopify-mcp \
  --clientId YOUR_CLIENT_ID \
  --clientSecret YOUR_CLIENT_SECRET \
  --domain your-store.myshopify.com

The server will exchange those credentials for short-lived tokens (valid for ~24 hours) automatically.

For legacy custom apps with a static shpat_ token:

claude mcp add shopify -- npx shopify-mcp \
  --accessToken YOUR_ACCESS_TOKEN \
  --domain your-store.myshopify.com

Important: Make sure you're installing shopify-mcp and not shopify-mcp-server — they're different packages with different argument names, and mixing them up causes errors.

Option B: Composio (paid, almost zero setup)

If terminal commands give you hives, Composio offers a hosted MCP that handles all the OAuth for you. You sign in once at composio.dev, connect your Shopify store through their UI, and they give you a single MCP URL to add to Claude Code. It costs money but saves about an hour of fiddling. For merchants running multiple stores or who don't want to manage credentials, it's worth it.

Either way, restart Claude Code after adding the server. Verify it's loaded with claude mcp list.

Step 5: Add Nano Banana for AI Image Generation

Nano Banana is the friendly name for Google's Gemini image generation models. Nano Banana 2 (Gemini 3.1 Flash Image) is the current default — it gives you 4K-capable output, character consistency across multiple generations, and crisp text rendering inside images, which is the part that finally makes AI usable for actual product photography.

You'll need a Gemini API key first. Go to Google AI Studio (aistudio.google.com), sign in with your Google account, and create an API key. Image generation requires billing to be enabled — text models are free, but image generation is pay-as-you-go. There's no monthly minimum.

Then add the MCP:

claude mcp add nano-banana \
  --env GEMINI_API_KEY=your_api_key_here \
  -- npx @rafarafarafa/nano-banana-pro-mcp

Restart Claude Code and you can now ask things like:

"Generate a hero image for the new linen curtain collection — natural morning light coming through a window in a Scandinavian living room, soft beige tones, 16:9 aspect ratio, 4K"
"Take this product photo and edit it to show the curtain on a different wall color"

The MCP supports text-to-image, image-to-image editing, and reference images for style transfer. Cost expectation for a working store: generating 50–100 images per month for product variations, lifestyle shots, and hero banners costs roughly $5–15.

Step 6: Add Google Search Console MCP

This is where the SEO loop closes. With Claude Code reading your live Search Console data, it can identify underperforming pages, propose fixes, apply them to your Shopify products and theme, and resubmit URLs for indexing — all in one session.

The most actively maintained option is mcp-gsc by AminForou. If you have uvx installed (the Python equivalent of npx), the simplest setup is:

claude mcp add gsc -- uvx --from git+https://github.com/AminForou/mcp-gsc mcp-gsc

You'll also need a Google service account with read access to your Search Console property:

Go to Google Cloud Console, create a project, and enable the Search Console API and (optionally) Web Search Indexing API.
Create a service account, generate a JSON key, and download it.
In Search Console, add the service account email as a user on your property with at least Restricted access.
Set the path to that JSON key as an environment variable when launching the MCP.

If you want a no-code path instead, Better Search Console (by Houtini) and Windsor.ai's hosted MCP connector both offer easier setup paths — Better Search Console is particularly nice because it syncs your data into a local SQLite database, which means Claude can run actual SQL queries against your historical data instead of being limited to the Search Console API's 1,000-row cap per request.

Three Workflows You Can Run on Day One

Setup is the boring part. Here's what makes the whole thing worth the hour you just spent.

Workflow 1: Bulk SEO Rewrite of Underperforming Products

Open Claude Code and type:

"Use the GSC MCP to find the 20 product pages on my store with the highest impressions but lowest CTR over the last 28 days. For each one, pull the current title and meta description from Shopify, look at the actual search queries those pages are appearing for, and rewrite the title and meta description to better match search intent. Don't publish anything — stage all changes as a list I can review."

What happens behind the scenes: Claude calls the GSC MCP to pull search analytics, calls the Shopify MCP to fetch each product's current SEO fields, uses its own reasoning to draft improved versions, and prints them in a table for you to approve. When you say "looks good, apply them," it pushes the changes through the Shopify MCP.

This is the kind of task that takes a human SEO an entire afternoon. Claude does it in about 4 minutes. The cost: roughly $0.30–$0.80 in Claude API usage for 20 products.

Workflow 2: Generate Lifestyle Hero Images for a New Collection

"I just created a new collection called 'Spring Linen.' Pull the 8 products in it from Shopify, read the product descriptions, then generate one lifestyle hero image for each product using Nano Banana — natural morning light, Scandinavian interior, consistent visual style across all 8 so they look like a campaign. Save them to a local folder and prepare a draft to upload them as the primary product images, but don't publish until I confirm."

Claude reads your products through the Shopify MCP, crafts descriptive prompts using the actual product attributes (color, material, dimensions), generates the images through Nano Banana with consistent style instructions, saves them locally, and stages the upload. Total cost for 8 high-resolution images: roughly $0.30–$1.20 depending on the resolution tier.

The thing that used to be impossible — keeping a consistent visual style across a multi-product set — is the exact thing Nano Banana 2's character/object consistency was built for. This is where it actually shows up in your bottom line.

Workflow 3: Daily SEO Health Check

Add a saved instruction to your Claude Code project — a CLAUDE.md file in your project folder — that says:

"Every morning when I open this project, run a Search Console health check: pull yesterday's clicks vs the previous 7-day average, flag any page that dropped more than 30%, check the indexing status of any new products added in the last 48 hours, and alert me if any sitemap submission has failed. Then propose specific fixes."

Now every time you start a Claude Code session, it does that automatically. This is the closest thing to having a junior SEO who never forgets and never takes a day off.

Cost Expectations (Realistic, Not Marketing)

For a small-to-mid Shopify store running this stack actively:

Claude Code (Anthropic) usage: $20–80/month depending on how often you run it. Most merchants land around $30–50.
Nano Banana (Gemini API): $5–25/month for typical product photography needs.
Google Search Console MCP: Free (uses your existing Google account).
Shopify Dev MCP: Free.
Shopify store MCP: Free if self-hosted, ~$20–50/month if using Composio.

So somewhere between $25 and $150 a month, all in, to replace tasks that used to take 10–20 hours a week of manual work or a part-time VA at €600–1200/month.

Safety Tips (Read This Before Running Anything)

A few rules I've learned the hard way working with merchants on this stack:

Never give an MCP write scopes you wouldn't trust an intern with on day one. Start with read-only, watch what Claude does, then expand permissions as you build trust. You can always add scopes later. You cannot un-delete a product collection.
Always use staging or draft mode for the first few weeks. Tell Claude explicitly to "stage everything as a draft and let me review before publishing." Most Shopify resources support draft states.
Keep a backup export of your products. Run Shopify's built-in product export weekly. Five minutes of insurance against a bad prompt.
Use a CLAUDE.md file in your project folder to document your brand voice, your no-go topics, your pricing rules, and any constraints. Claude reads this on every session start. Without it, you'll get inconsistent output that drifts from your brand over time.
Audit the MCPs you install. They're running on your machine with access to your store. Stick to actively maintained projects from credible authors. Check the GitHub stars, the last commit date, and whether the maintainer responds to issues.

What This Really Changes for Your Store

The honest pitch isn't "AI replaces your team." It's that the boring 70% of running an online store — the SEO updates, the image refreshes, the inventory sweeps, the description rewrites — becomes something you supervise instead of something you do. You get your weekends back, and the parts of the business that actually need a human (positioning, supplier relationships, customer empathy, brand decisions) get more of your attention because they're not being crowded out by busywork.

That's not a future-of-work prediction. That's a stack you can install this afternoon.

FAQ

Do I need to know how to code? No. Everything in this guide uses copy-paste terminal commands. You need to be comfortable opening a terminal and pasting a line — that's it. Claude Code handles the technical parts.

How much does the full stack cost per month? Between $25 and $150 depending on usage. Most small Shopify stores land around $50–80/month. That replaces 10–20 hours of manual work weekly.

Is it safe to give Claude Code access to my live store? Yes, with precautions. Start with read-only scopes, use draft mode for all changes, and keep weekly product backups. Expand permissions gradually as you build trust.

What Shopify plan do I need? Any plan that supports custom apps — which is all paid plans. The Dev MCP works with any plan. The store-management MCP needs the custom apps feature.

Can I use this with multiple Shopify stores? Yes. Add a separate MCP instance for each store with different credentials. Claude Code can switch between them in the same session.

What if an MCP server stops working? MCP servers are community-maintained. If one breaks after a Shopify API update, you can temporarily remove it with claude mcp remove [name] and re-add it when updated. The Shopify Dev MCP is official and maintained by Shopify.

Need Help Setting This Up?

If you got partway through this guide and hit a wall — Windows path issues, OAuth headaches, MCP servers refusing to load — that's normal. The first install is the hardest.

At areza.digital, we set up exactly this stack for European Shopify merchants as a productized service. We handle the auth, the credential rotation, the CLAUDE.md brand setup, and the first three automation workflows tailored to your store. You walk away with a working system and a 30-minute training call so you can run it yourself.

Book a 30-minute discovery call →

Written by Nikita Janochkin, founder of areza.digital. Sources: official Shopify Dev documentation, Anthropic Claude Code docs, Google AI Studio, AminForou/mcp-gsc, GeLi2001/shopify-mcp, ConechoAI/Nano-Banana-MCP, Composio, and direct testing on live Shopify stores. Last updated April 8, 2026.

Stop losing leads to a slow website

Book a free friction audit and see exactly where your website is leaking money.

Book a call →