Connect Shopify to Claude or ChatGPT with MCP: What Works Today

The question arrives in my inbox most weeks in the same shape: can I connect my Shopify store to Claude, or to ChatGPT, and just ask it things. Yes, with a catch that most of the guides skip. Shopify’s official servers are built for the people shopping at your store and the people coding for it. The one you are picking up the laptop for, the one that reads your orders, is not official, and the honest post says so before the setup instructions.

Shopify's three official MCP servers and what each sees: the Storefront server at /api/mcp for shoppers and agents, catalogue, cart and policies; the Customer Accounts server for a logged in shopper's orders and returns; the Dev server for coding tools, docs and GraphQL schemas. Admin data, orders, inventory and customers, is reached through Sidekick, a community server or a custom one

The three official servers

Storefront MCP. Every Shopify store already serves it at https://yourstore.myshopify.com/api/mcp. No app, no setup. It exposes what a shopper can see: product search, collections, cart operations, shipping and returns policies, and the start of a checkout. It exists so that AI shopping agents can buy from you without scraping, and it is the plumbing behind the agentic commerce and llms.txt rollout I wrote up in May. For an owner, the job is to keep it healthy: clean titles, correct availability, policies that say what you mean. Listing products on ChatGPT covers that side.

Customer Accounts MCP. The same idea for a logged in customer: their orders, tracking, returns. Authentication runs through Shopify’s customer account flow. Useful if you are building a support agent that acts on behalf of a shopper. Not useful for asking about your business.

Dev MCP. npx -y @shopify/dev-mcp@latest gives a coding assistant Shopify’s documentation and GraphQL schemas, so Claude Code or Cursor stops inventing API fields. Shopify shipped it in the Winter ‘26 edition and open sourced the AI Toolkit around it in April 2026. If anyone on your team touches theme code, this is worth five minutes. If nobody does, skip it.

What you actually wanted: the admin

Orders, inventory, customers, refunds, the numbers. Three routes, in the order I would try them.

Five questions to ask Sidekick before connecting anything: most refunded products, conversion before and after the theme update, yesterday's sales by channel, all answered inside the admin; which products the winback flow sold and Ads purchases against orders by day need Klaviyo or GA4 too, which is what MCP is for

Sidekick, first. It is inside the admin on every plan and it already answers the questions people imagine asking Claude: which products had the most refunds last month, how did conversion rate move after the theme update, which code was used most. Before connecting anything, open Sidekick and ask it the three questions you most want answered. For a lot of stores that is the end of the project, and the analytics walkthrough shows where each number lives when you want to see it without asking.

A community Admin MCP, second. Several exist on GitHub. They run locally, take a custom app access token from your store, and expose orders, products, customers and inventory to Claude Desktop or Cursor. They work. Two cautions: they are maintained by individuals and can break when Shopify’s API version rolls, and most of them ship with write tools enabled. If you use one, create the custom app with read scopes only, and nothing else.

Before a community admin server touches the store: create one custom app per assistant under Develop apps, scope it to read orders, products, inventory and customers only with no write scope, store the token on a server, revoke it when the assistant is removed

A custom server, when it has to be reliable. For a store that wants the owner, the ops lead and the support team all asking the admin questions every day, I build a small read only MCP server on the Admin GraphQL API, scoped to the queries the team actually asks, with the token on a server rather than a laptop. It is the same thing I did for a point of sale that had no official server at all, and it took a weekend to a working version. The write up covers what the owner asks it and what to be careful about.

Connecting the official ones

For the Storefront server in Claude Desktop, add a remote MCP server with the URL https://yourstore.myshopify.com/api/mcp. No token. Ask it “what do you sell under $50” and you are seeing your store as an agent sees it, which is a useful exercise in itself.

For the Dev MCP in Claude Code: claude mcp add shopify-dev -- npx -y @shopify/dev-mcp@latest. Then ask it to write a GraphQL query for orders with a particular tag and watch it get the field names right.

Both of these are the same shape as Google’s servers for Analytics and Ads, and if those are already connected, the interesting questions are the ones that cross systems: which products sold yesterday, and which campaigns sent the traffic.

What I would do

Ask Sidekick first. Connect the Storefront server once to see your store through an agent’s eyes and fix what looks wrong. Skip the community admin servers unless you are comfortable creating a read only token and rotating it. If the team would use a real one every day, that is a build I do, and the plain language version of what an MCP server is, for the owner who has to approve it, is what is an MCP server, for store owners.

If Shopify ships an official admin server after this is published, this page will say so at the top.

Frequently asked questions

Does Shopify have an official MCP server?
Three, as of October 2026. The Storefront MCP at yourstore.myshopify.com/api/mcp, on every store with no setup, which lets an agent search products, manage a cart and read policies. The Customer Accounts MCP for a logged in shopper's orders and returns. And the Dev MCP, installed with npx @shopify/dev-mcp, which gives coding tools Shopify's documentation and GraphQL schemas. All three are public or developer facing. None of them reads your admin.
Can I ask Claude about my Shopify orders and inventory?
Not through an official Shopify server yet. The storefront server sees what a shopper sees. For admin data you have three routes: Sidekick inside the Shopify admin, which answers most owner questions already; a community Admin MCP server running on your own access token, which works and may break when the API changes; or a small custom server on the Admin GraphQL API, which is what I would build for a store that wants this reliably.
What is the Storefront MCP endpoint for?
Agents that shop. ChatGPT, Perplexity and the rest can call it to search your catalogue, check a policy and build a cart without scraping your pages. It is why the agentic commerce work on this site matters: it is the door AI shopping assistants walk through. For a store owner it is something to keep healthy, not something to query.
Is the Dev MCP useful if I am not a developer?
Only if you or someone on your team writes theme code, Liquid, or app integrations with an AI coding tool. It makes Claude Code or Cursor accurate about Shopify's APIs instead of guessing. For an owner who wants answers about the store, it does nothing.
Is it safe to connect an AI to my Shopify admin?
With a read only access token scoped to the data you want answered, yes, and that is the only way I would do it. Never give an assistant a token with write scopes to orders, products or customers. Everything I build is read only by default, runs under your own token, and the model sees only the result of the query it ran.
What questions does Sidekick already answer?
Most of the reporting ones: which products had the most refunds last month, which discount code was used most, how conversion rate changed after a theme update, what sold yesterday by channel. It is inside the admin on every plan, needs no setup and is the first thing to try before connecting anything else.
Nikhil Sharma

Nikhil Sharma

I run data and operations for a DTC brand on Shopify, consult on Meta and Google Ads, and build AI automations for ecommerce teams. I write about what I build and run.

Book a call More about me