Shopify MCP is the set of servers that let an AI tool such as Claude read and act on a Shopify store through the Model Context Protocol (MCP). There is no single “Shopify MCP”. There is one for merchants who manage their own store from a chat, one for developers, and a group of agent servers that let shopping assistants search a catalog and build a cart. This guide is for a merchant or eCommerce lead, and it follows Shopify’s documentation as of 29 October 2026.
We are noprob agency and we sell the Shopify Launch Sprint. Our team works every day with AI agents built on Claude Code. The plan: what MCP is in two sentences, then every Shopify server one by one, what you need to connect them and the permissions we would give.
What is MCP, in two sentences?
MCP is an open standard that lets an AI app ask an outside system to run a tool, in the same way every time. MCP’s own introduction compares it to a USB-C port: one standard plug instead of one cable for every pair of devices.
A model on its own only writes text. To read your orders or change a price it needs a tool, and a tool needs a connection. With MCP, the system publishes a server that lists its tools (search products, get an order, update a product) and any compatible AI app can call them. Anthropic created MCP and in December 2025 moved it to the Agentic AI Foundation, under the Linux Foundation, which it co-founded with Block and OpenAI.
What MCP does not do: decide the rules. It standardizes the plug, not who may call what, which actions need your confirmation or what happens when the model gets it wrong. That part stays yours. The MCP specification itself says there should always be a human who can deny a tool call.
Which Shopify MCP servers exist today?
Shopify MCP comes in three families: the connector for your own store, the developer tools, and the agent servers for assistants that serve shoppers. Here is the map, with what each does, who uses it and what you need.
| Server | What it does | Who uses it | What you need |
|---|---|---|---|
| Shopify connector (Claude and other AI tools) | Lets you manage your own store from a chat (products, collections, inventory, discounts, orders, analytics). | Merchants and eCommerce leads. | Access to the Shopify admin, an account with the AI tool and your approval of the data access it requests. |
| Dev MCP and AI Toolkit | Give an AI coding assistant Shopify’s current documentation, API schemas and code checks. | Developers and agencies. | Node.js and one command or plugin install; Dev MCP needs no authentication. |
| Storefront Catalog MCP | Searches and looks up one store’s products and physical locations; no tool changes anything. | Builders of shopping assistants and on-site agents. | As a merchant, Shopify’s page describes no setup; the agent must present an agent profile. |
| Global Catalog MCP | Searches products across Shopify merchants from one endpoint. | Agents that compare stores. | As a merchant, no opt-in step is described; what you control is the quality of your product data. |
| Cart MCP and Checkout MCP | Build a cart over several turns (Cart MCP), then turn it into a checkout (Checkout MCP). | Agents shopping for a buyer. | The agent needs a profile, and Checkout also needs a token or a signed request. |
| Ask MCP and Order MCP | Ask answers questions on policies such as returns and shipping, and Shopify marks it as a draft capability; Order reads the current state of an order, only for orders placed through the agent. | Support and order-status assistants. | The agent’s profile, and for Order a token with the order-read scope. |
| Customer Accounts MCP | Handles tasks for a logged-in customer, such as order status and account preferences. | Developers of assistants for existing customers. | A custom domain, approval for protected customer data and customer account login. |
The name “Storefront MCP” is the one that confuses people. It used to be one server at your store’s /api/mcp address. Shopify’s migration page says its catalog and cart tools were removed and replaced by the same capabilities on the Universal Commerce Protocol (UCP) at /api/ucp/mcp, so a guide that still shows the old endpoint is out of date.
UCP is the commerce vocabulary those servers speak. ucp.dev presents it as the common language for platforms, agents and businesses, co-developed by industry leaders including Google, Shopify, Etsy, Walmart, Amazon and Stripe, and covering catalog search, cart, checkout, identity and orders. MCP is how an AI app talks to the server; UCP is what the server says about shopping. The wider picture of agents that buy is in agentic commerce for Shopify.
What can a merchant do with the Shopify connector for Claude?
With the connector you can run daily catalog and reporting work from a chat, with write access to products, collections, inventory and percentage-off discount codes, and read-only access to orders, customers and analytics. That split comes from Shopify’s help center, and it is a good one.
Examples of work that fits (the examples are ours, the capabilities are Shopify’s):
- Find products with no description and draft the text, which you review before it goes live.
- List variants with low stock and propose the inventory changes. The connector reads the current quantity first, so an update stops if the count has changed.
- Ask an analytics question, such as which products sold most in the last 30 days.
- Create a percentage-off discount code for a campaign.
- Work on an unpublished copy of a theme, never on the live one.
What it cannot do, even with write access: refunds, order cancellation or capture, marking an order paid, returns, gift cards, and changes to plan, payments, taxes or domains. It cannot edit or delete your live theme, and it asks you to confirm before publishing a theme or making certain Admin API changes. Those limits are a feature. They are the reason we are comfortable suggesting it to a store owner.
Setup is short: install it from Claude’s connector directory, not from the Shopify App Store, use a desktop browser, and approve the access in your Shopify admin. What it can do depends on the permissions of the staff account you approve it with. Shopify’s help page does not call it MCP; the connector address on Claude’s listing ends in /mcp. Shopify says other AI tools can connect too, among them ChatGPT, Gemini and Microsoft Copilot. The Claude side, including Claude Code, is in Claude and Claude Code on Shopify. Shopify’s own assistant inside the admin is covered in Shopify Magic and Sidekick.
What do Dev MCP and the AI Toolkit do?
Dev MCP and the AI Toolkit give an AI coding assistant Shopify’s current documentation and checks, so it writes Liquid, GraphQL and extensions against the real schema instead of its memory. They are for the person building the store, not for the person selling in it.
The Dev MCP server runs on your computer and needs no authentication. In Claude Code the page gives one command: claude mcp add --transport stdio shopify-dev-mcp -- npx -y @shopify/dev-mcp@latest. The AI Toolkit is the plugin route: documentation, API schemas, validation of GraphQL, Liquid and extensions, and store-management tasks that run through Shopify CLI, when you choose to run them. Shopify is explicit that it does not remove the need to review and test the result.
For a merchant the useful question is a different one: if you hire a developer or an agency, do their agents work from current Shopify documentation, and who reviews what the agent writes? In our Custom AI System, “development with AI agents” is one of the six superpowers, and the review stays with people.
How do shopping agents use Storefront, Cart, Checkout and Ask MCP?
Shopping agents use the catalog servers to find products, Cart and Checkout MCP to prepare the purchase, and Ask and Order MCP for policy and order questions, and Shopify limits each agent by how well it identifies itself. Shopify’s page on auth and rate limiting describes three levels: an agent with a Dev Dashboard credential, one that signs its requests, and an anonymous one. Only the first can complete purchases or read orders, and only with the right permission, and Checkout is limited more strictly than Cart.
The Storefront Catalog server is read-only in practice: its tools are search and lookup, at an address on your store domain, and the page describes no merchant setup. The Global Catalog does the same across merchants. Cart MCP builds carts and does not take payment; Checkout MCP hands the buyer to your storefront to finish unless a trusted agent has been granted permission to complete it. The Customer Accounts server needs a custom domain, approval to use protected customer data and an OAuth login.
For you, the work is data, not connection: complete attributes, accurate prices and stock, readable policies. That is the same work as in GEO for Shopify, and it is easiest during a migration, when every product is touched anyway: see our guide to migrating to Shopify. On Nazzi’s store we published a sitemap for agentic discovery. If you want your own assistant on top of these servers, the choice between chatbot, workflow and agent is in AI agents for Shopify.
What do you need to connect a Shopify MCP server?
What you need depends on the server, but it is always an account, a permission and a decision about who owns the connection. In order:
- For the connector: a Shopify staff account with the right permissions, an account with the AI tool and a desktop browser for the first setup.
- For Dev MCP and the AI Toolkit: Node.js (the Toolkit asks for version 18 or higher) and one command or plugin install in Claude Code, Codex, Cursor or VS Code.
- For agent servers: an agent profile hosted at a public address, credentials from the Dev Dashboard for the highest level and, for checkout, a token or a signed request.
- For Customer Accounts: a custom domain, approval for protected customer data and customer account login configured in the app.
- For all of them: one named owner, who reviews what the AI changed and knows how to revoke the access.
What are the risks, and which permissions should you give?
The main risks are an AI changing data you did not mean it to, an AI following instructions hidden in content it reads, and access that outlives the person who set it up. The answer to all three is the same: least privilege, read-only first, and confirmation on every action that matters. Our rules:
- Start read-only. In the connector, orders, customers and analytics already are. Keep it that way for the first weeks and use it to ask questions.
- Give the connection its own staff account with only the permissions the job needs. The connector’s reach follows the staff account it uses.
- Write only where you can undo. Product drafts and unpublished themes yes; prices on the whole catalog in one go, no. Shopify notes that a large batch changes items one at a time and can stop partway.
- Keep money out of reach. Refunds, cancellations and returns are blocked in the connector. If you build your own, do the same and require a human approval, as the MCP specification recommends.
- Watch the update calls. Cart MCP says update_cart replaces the cart’s full state and removes any field you omit; Checkout MCP says the same of update_checkout. An agent that sends half a cart deletes the other half.
- Treat everything the agent reads as untrusted. Reviews, product text and customer messages can carry instructions. OWASP lists prompt injection as a top risk for LLM apps and recommends least-privilege access and human approval for high-risk actions. The full picture is in AI agent security for Shopify.
- Check the output, and know where the data goes. Shopify’s considerations page says you are responsible for wrong changes and that the data you share is processed by the AI tool’s company under its own terms.
Our method for a first project is a ladder: questions only, then drafts you approve, then small writes with a log, and money actions never delegated.
Where should you start this week?
Start with the connector in read-only use, and fix your product data in parallel. In this order:
- Pick one owner and one dedicated staff account with limited permissions.
- Connect the Shopify connector and ask five questions about sales, stock and products. Change nothing.
- Let it draft descriptions for ten products and review them yourself.
- Check your product data and policy pages, because every agent server reads them.
- Ask your developer or agency which of Dev MCP, the AI Toolkit or another tool they use and how they review the output.
How does this fit with the Shopify Launch Sprint and the AI Accelerator?
In the Shopify Launch Sprint, the Custom AI System leaves the store ready for AI agents, with clean and mapped data. “Ready for AI agents” and “data mapped in detail” are two of its six superpowers, and they are what the catalog, cart and Ask servers read. The Sprint takes 14 days, with go-live between day 15 and 30 and 90 days of support. If you are still on another platform, start from our guide to migrating to Shopify. If your store is already on Shopify, the AI Accelerator is the lighter route.
Frequently asked questions
What is MCP in Shopify?
MCP (Model Context Protocol) is an open standard that lets an AI app such as Claude call tools on an outside system in a uniform way. In Shopify it means servers that let an AI tool read and, with your approval, change your store, or let shopping assistants search a catalog and build a cart.
Is Shopify MCP free?
Shopify’s help center says there is no extra Shopify fee for the connector: you pay your Shopify plan and the subscription of your AI tool. Shopify’s pages for the developer and agent servers mention no fee, but any assistant you build pays its own AI model costs.
Which Shopify MCP server do I need?
If you run a store and want to manage it from a chat, the Shopify connector for Claude or another AI tool. If you or your agency write code, Dev MCP and the AI Toolkit. If you build an assistant that serves shoppers, the catalog, cart, checkout, Ask, order and customer account servers.
Can I use Shopify MCP with Claude or Claude Code?
Yes. Merchants add the Shopify connector from Claude’s connector directory and approve access in the Shopify admin. Developers can add Dev MCP to Claude Code with one command, or install the AI Toolkit plugin. Both are covered above.
Is it safe to connect an AI tool to my Shopify store?
It is safe if you control the permissions. Start read-only, use a staff account with only the access the job needs, keep refunds and other money actions out of reach, and review every change. Shopify’s connector already blocks refunds, order cancellation and live theme edits.
Does the old Storefront MCP endpoint still work?
Not for catalog and cart. Shopify’s migration page says the catalog and cart tools on the old /api/mcp endpoint were removed and replaced by UCP-based tools at /api/ucp/mcp. The Customer Accounts MCP server is unchanged.
The Sprint has a public price by revenue: €5,800 under €2M, €8,800 from €2M to €20M, €12,800 above €20M, VAT excluded, in 4 installments.
The Shopify Launch Sprint moves your store to Shopify in 14 days with our Custom AI System: a fixed calendar, a public price and 4 written guarantees.
Antonio Manitta
Founder & eCommerce Manager at NoProb Agency
For over 7 years I have helped fashion, supplements, and DTC brands scale their eCommerce. I work in long-term partnership with three brands at a time, building stable systems that compound.
About NoProb →
