Shopify MCP Server

The Shopify MCP server is how an AI agent gets a live line into a store: products and variants, collections, inventory levels, orders, customers and discounts, through the Admin GraphQL API. It is the connector most ecommerce agents start with, because almost every question about the business begins with something the store already knows. Shopify builds and maintains it, and it is listed in Claude's connector directory, so there is no config file to write: you add it from the directory and authorise it in your Shopify admin.

It also exposes the API directly — run a query, run a mutation, introspect the schema — so an agent is never boxed in by the packaged tools. That is the part most write-ups miss, and it is what separates this from a read-only reporting bridge.

Which Shopify MCP server are you actually looking at?

Four different things get called the Shopify MCP server, and they do not do the same job. Picking the wrong one is the most common way to lose an afternoon.

• The Shopify connector for Claude — the one described on this page. Built by Shopify, listed in Claude's directory, 25 tools, reads and writes your live store. One click to connect, nothing to install.
• Shopify Dev MCP — for people building on Shopify. It searches Shopify's documentation and introspects the Admin API schema. It never touches your store's data.
• Storefront MCP — the customer-facing one, served from your own domain. It powers shopping experiences for buyers, not internal analysis.
• Community admin servers — npm packages that run locally against the Admin API with a token you create yourself. Smaller tool sets, more setup, and nobody maintains them for you. Worth it only if the server has to live inside your own infrastructure.

Shopify MCP server vs the Shopify Admin API

The MCP server is not a replacement for the Admin API. It is the Admin API, wrapped so that a model can find its way around it.

• Discovery. The API assumes you already know which resource and which field you want. The MCP server ships graphql_schema and search_docs_chunks, so the agent looks up the shape of the data before asking for it.
• Authentication. The API needs an app, scopes and a token you manage. The connector is an OAuth flow you approve once in your Shopify admin.
• Guard rails. Twenty-five packaged tools cover the common operations with sane inputs. When they do not fit, graphql_query and graphql_mutation leave the whole API open.
• Who writes the code. With the API, you do. With MCP, the agent composes the call and you read the result.

The trade-off is the same one every MCP server makes: convenience at the cost of a narrower, opinionated surface — with, here, an escape hatch back to raw GraphQL.

What you can ask once Shopify is connected to Claude

Reading the store

• “Which products sold most units last month, and which of them are now under ten units in stock?”
• “Pull order 1043, its line items and its fulfilment state.”
• “List the collections a product sits in, and what else is in them.”
• “How many customers ordered more than twice this quarter?”

Changing the store

• “Draft ten products from this spreadsheet, leave them all in draft status.”
• “Move every product tagged archive-me to archived.”
• “Create a 15% code for the summer collection, valid two weeks, one use per customer.”
• “Set stock for these six SKUs at the Rotterdam location.”

Writes land on the live store. Shopify has no staging copy of your catalogue, so treat a write prompt the way you would treat a bulk edit in the admin: say what you want changed, then read back what it did.

What the Shopify MCP server cannot answer

Everything on this page stops at the edge of Shopify. The server knows orders; it does not know what an order cost you to acquire. Ask it for blended ROAS, customer acquisition cost, contribution margin, or how a Meta campaign performed against an email flow, and there is nothing in the Admin API to answer with.

That is not a flaw in the connector, it is the shape of the data. Shopify is one source. The questions operators actually ask on Monday morning span five or six.

Where Polar fits next to it

Polar sits one layer under the agent: 45+ sources reconciled into a dedicated warehouse, with a semantic layer where CAC, LTV, net revenue and contribution margin are defined once and used everywhere. Connect both and the division of labour is clean — Shopify answers questions about the store, Polar answers questions about the business, and neither is guessing at the other's numbers.

Frequently asked questions

Is there an official Shopify MCP server?

Yes. Shopify publishes and maintains the connector listed in Claude's directory, alongside a developer-facing server for its documentation and API schema, and a storefront server for buyer-facing experiences. The three serve different audiences.

Do I need a developer account or a custom app?

Not for the Claude connector. You add it from the directory and authorise it from your Shopify admin. Community servers are the ones that ask you to create an app and manage credentials yourself.

Can it change my store, or only read it?

It writes. Products, collections, inventory, discounts and arbitrary GraphQL mutations are all in the tool list. Read the plan back before approving anything that touches many records at once.

Does it work with more than one store?

Yes. switch-shop moves between the stores you have connected, inside the same conversation.

Why can't it tell me my blended ROAS?

Because ad spend does not live in Shopify. Blended ROAS needs the store joined to every ad platform, on one definition of revenue — which is the job of a semantic layer, not of a platform connector.