Shopify MCP Explained for Merchants: Which One Does What

Search "Shopify MCP" and you'll get a developer tutorial, an app listing, a guide to connecting Claude, and an article about AI shopping agents, all on one page of results and all using the same three letters for different things. No wonder it's confusing.

Here's the short version. MCP is a way for an AI assistant to call tools on a system instead of guessing from a web page. On Shopify, there are three separate groups of those tools, built for three different users:

1. Shoppers' AI agents, searching and buying from your store
2. You, managing your store from an assistant like Claude or ChatGPT
3. Developers, building apps and themes with AI coding tools

Once you know which group a page is talking about, the rest falls into place. If you want the background on the protocol itself, start with MCP for Shopify stores.

Group 1: MCP for shoppers' AI agents

This is the side Shopify calls agentic commerce. When a shopper asks an AI assistant to find or buy something, the assistant needs a reliable way to search products, build a cart, and check out. Shopify provides that as a set of tools, described in its agents documentation:

• Catalog search, across all Shopify merchants or scoped to one store
• Cart, to build and change a cart
• Checkout, to turn a cart into a checkout. Trusted agents can complete the purchase directly, and others hand the buyer to your storefront.
• Order, to fetch the current state of an order

On your own store, these sit at standard paths that Shopify provides: a discovery endpoint at /.well-known/ucp and an MCP endpoint at /api/ucp/mcp. We walk through how an agent uses them in UCP and MCP endpoints in agents.md.

Who uses it: AI shopping assistants, on behalf of buyers. Not you.

What you do about it: You don't build or host anything. What's in your hands is the quality of what those tools return. An agent asking for your product gets back whatever your catalog holds: titles, variants, prices, availability, identifiers, policies. Thin data in, thin answer out. Which channels can reach your store is a setting in your admin, covered in which AI channels to turn on.

Group 2: Connectors for you, the merchant

This is the newer one, and the one most merchants are actually looking for when they type "connect Claude to Shopify."

Shopify maintains official integrations for ChatGPT, Claude, Gemini, Grok, Perplexity, Muse, and Microsoft Copilot. They work on the same idea as everything above: the assistant gets a set of named actions it can call. The difference is whose side it's on. These run with your permissions, against your store, for your work.

With the Claude and ChatGPT integrations, the actions include searching products, reading inventory by location, listing recent orders, running analytics questions, and, if you approve write access, creating products, setting inventory, and creating percentage-off discount codes.

Who uses it: You and your staff.

What you do about it: Install the integration from inside your AI tool, approve access in your admin, and start with read-only. Setup guides: Claude and ChatGPT. They aren't in the Shopify App Store, which is part of why they're hard to find.

Group 3: MCP for developers

Two tools here, and neither is meant for running a store day to day.

The Dev MCP server. Part of the Shopify AI Toolkit. It connects AI coding tools like Claude Code, Cursor, and VS Code to Shopify's developer documentation and API schemas, so the code they write is accurate. It runs locally and doesn't need authentication, because it's reading Shopify's public developer resources, not your store.

The Shopify CLI connector. For a developer working on a store from their own machine. It authorizes a local AI coding tool to reach the Shopify API for one store, so they can build and test an app or theme against real data. Shopify's help page says plainly that it isn't a chat interface and isn't the right fit for managing a store from a chat.

Who uses it: Developers and agencies.

What you do about it: Nothing, unless you write code. If a tutorial has you installing Node packages and editing a JSON config file to "manage your store with AI," you've wandered into this group by mistake. Go back to Group 2.

The cheat sheet

What about third-party MCP apps and DIY servers?

Before the official integrations existed, the only way to get an assistant into your store was a third-party MCP server: create a custom app, generate an Admin API token, paste it into a config file. A lot of search results still describe that route, and there are apps in the App Store that package it up.

Shopify's current guidance is to use the official integration for each AI tool and not to create an API access token or build your own integration to connect one. For most merchants that settles it.

A third-party connector can still make sense if you need something the official route doesn't offer, such as an audit log or managing several stores from one place. If you go that way, look at which permissions it requests, where your data passes through, and how you'd revoke it. We cover the trade-offs in is it safe to connect AI to your Shopify store?

One store, two audiences, same data

Here's what ties the first two groups together, and why it matters beyond terminology.

When a shopper's agent calls your store's tools, it reads your product data. When your own assistant answers "what's in stock?", it reads the same product data. Different doors, same room.

A catalog with clear titles, complete variants, real identifiers, and policies written as content gives good answers through both. A catalog with gaps gives you a vague assistant and gives shoppers' agents a reason to recommend someone else. That second cost is invisible, because a recommendation you didn't get doesn't show up in any report.

You can see what AI assistants can read from your storefront with the free AI readiness checker. It takes about a minute.

For the wider map of every way to use AI in your store, including Sidekick and Flow, see every option compared.