Shopify MCP is the set of MCP servers and connectors that Shopify makes so an AI assistant can work with Shopify. Store owners use the Shopify connector for Claude to read and manage a store from a chat. Developers use the Dev MCP server. Teams building shopping agents use Shopify's UCP (Universal Commerce Protocol) servers.
Shopify makes all of them (Shopify Help Center and Shopify developer docs, checked 7 Oct 2026). Depra AI makes none of them. It needs nothing installed in your store, and it never changes a store. The one thing in this guide that Depra AI makes is the read-only Depra AI MCP server.
MCP (Model Context Protocol) is an open-source standard that connects AI applications to outside systems. Its makers compare it to a USB-C port for AI apps (Model Context Protocol, checked 7 Oct 2026). This guide covers each Shopify server and its setup lines. It then pairs Shopify's connector with the Depra AI MCP server, so one assistant reads your catalogue and your AI visibility data together.
| Server or connector | Who it is for | What it does | What it needs |
|---|---|---|---|
| Shopify connector for Claude | Store owners and marketers | Reads a store you own from a Claude chat, and changes it once you approve write access | Install from Claude, then review the data access it requests in your Shopify admin |
| Shopify AI Toolkit, with the Dev MCP server | Developers | Developer docs and API schemas, code checks, and store tasks through Shopify CLI | Node.js 18 or higher and a supported AI tool |
| UCP servers (Catalog, Cart, Checkout, Order) | Teams building shopping agents | Product search, carts, checkouts and order status for buyers | An agent profile on every request |
| Storefront MCP (older) | No longer the route | Its catalog and cart tools were removed | Move to UCP |
What is Shopify MCP?
Shopify MCP is not one server. People use the name for three things Shopify makes: a chat connector for store owners, a developer toolkit, and commerce servers for shopping agents.
An MCP server gives an AI app, such as Claude Code or Cursor, a list of tools it may call. A connector is Claude's name for an MCP server that you add in claude.ai (Claude Code docs, checked 7 Oct 2026).
Shopify's help centre says to use the official Shopify integration for each AI tool, not an API access token or a build of your own (Shopify Help Center, checked 7 Oct 2026). For other vendors, see the best MCP servers for SEO and marketing.
Which MCP servers does Shopify offer today?
Shopify offers a chat connector for Claude, the AI Toolkit with its Dev MCP server, and UCP servers for shopping agents. This guide covers those three. Shopify also lists integrations for ChatGPT, Perplexity and Muse, which this guide leaves out (Shopify Help Center, checked 7 Oct 2026). The older Storefront MCP server is partly retired.
Shopify connector for Claude: for store owners
The connector lets you create or manage a Shopify store from a Claude chat. It works with products, collections, inventory, discounts and analytics, and it reads orders and customers without changing them (Shopify Help Center, checked 7 Oct 2026).
It is not read-only. Once you approve write access, it can create and edit products and collections, set stock counts, create discount codes, and change pages, blogs and menus. Even then, it cannot edit your live theme, issue refunds or change store settings (Shopify Help Center, checked 7 Oct 2026).
Claude's directory lists it as made by Shopify, added in April 2026 (Claude connector directory, checked 7 Oct 2026).
Shopify AI Toolkit and the Shopify Dev MCP server: for developers
The Shopify AI Toolkit gives an AI coding tool Shopify's developer docs and API schemas, the lists of fields an API accepts. It also checks code. It can prepare store tasks that run through Shopify CLI, a tool run by typing commands, when you choose to run them. It installs as a plugin, as agent skills (instruction sets an AI tool can load), or as the Dev MCP server. That server runs on your own computer and needs no sign-in (Shopify AI Toolkit, checked 7 Oct 2026).
Store access is separate. The Shopify CLI connector lets an AI tool on your computer read and change one store's data, within the access you grant (Shopify Help Center, checked 7 Oct 2026).
UCP servers: for teams building shopping agents
UCP (Universal Commerce Protocol) is the standard Shopify uses for AI agents that shop for buyers. Its UCP servers cover product search, carts, checkouts and orders (Shopify agentic commerce docs, checked 7 Oct 2026).
The Storefront Catalog MCP server searches one store's products at https://{storedomain}/api/ucp/mcp. Every request must carry an agent profile, a web address that identifies the agent (Storefront Catalog MCP, checked 7 Oct 2026).
What happened to the Shopify Storefront MCP server?
Shopify removed the catalog and cart tools of the Storefront MCP server at https://{shop}/api/mcp and points developers to UCP. For AI chat on a store without code, Shopify suggests its Inbox agent (Migrate from Storefront MCP, checked 7 Oct 2026).
How do you connect a Shopify MCP server to Claude Code or Cursor?
You connect it with one command in Claude Code, or one setting in Cursor. The lines below are copied from Shopify's documentation, which asks for Node.js 18 or higher, the software these tools run on (Shopify AI Toolkit, checked 7 Oct 2026). It also covers Codex and VS Code.
The first four entries below are developer tools and do not connect a store. The fifth, the connector for Claude, is the one for store owners.
- Claude Code, as a plugin. Shopify recommends this install. In your terminal, the window where you type commands, run: claude plugin install shopify-ai-toolkit@claude-plugins-official
- Claude Code, as the Dev MCP server. Run this line, then restart Claude Code: claude mcp add --transport stdio shopify-dev-mcp -- npx -y @shopify/dev-mcp@latest
- Cursor, as a plugin. In Cursor Chat, type: /add-plugin shopify
- Cursor, as the Dev MCP server. Go to Cursor > Settings > Cursor Settings > Tools and MCP > New MCP server. Add this configuration, save it and restart Cursor: {"mcpServers": {"shopify-dev-mcp": {"command": "npx", "args": ["-y", "@shopify/dev-mcp@latest"]}}}
- The Shopify connector for Claude. Install it from its page in Claude, not from the Shopify App Store. Your Shopify admin then shows the data access it requests. Read it before you approve it (Shopify Help Center, checked 7 Oct 2026). Connectors added in claude.ai also appear in Claude Code when you log in with a claude.ai account (Claude Code docs, checked 7 Oct 2026).
How do you pair Shopify's MCP with the Depra AI MCP server?
Add both to the same assistant. Shopify's connector answers questions about your catalogue. The Depra AI MCP server answers questions about AI answers: the brands named, the sites cited and the products ChatGPT shows.
The steps use Claude Code, because the Depra AI MCP server cannot connect to claude.ai on the web yet.
The Depra AI MCP server has 24 read-only tools and is included on every plan while a trial or subscription is active. None of its 24 tool descriptions mentions a store catalogue. One tool, get_project_profile, returns the products field of your brand profile. So the catalogue comes from Shopify.
We wrote this pairing from each side's documentation and have not run it end to end.
- Step 1. Add the Shopify connector in Claude. Shopify says a connected tool can read data but not change it unless you approve write access (Shopify Help Center, checked 7 Oct 2026). The checks below only read, so they do not need write access.
- Step 2. In Depra AI, open API Keys in the sidebar, under Company, and click Create token. A personal access token is a secret key that lets an app sign in. A token can read only the projects its creator can access, so treat it like a password.
- Step 3. Add the Depra AI MCP server at https://api.depra.ai/mcp, which takes the token as an Authorization: Bearer header. On macOS or Linux, add export DEPRA_TOKEN=YOUR_DEPRA_TOKEN to your shell profile, with your token in its place, and open a new terminal. Then run the command from the Claude Code tab on the API Keys page: claude mcp add --transport http --scope user depra https://api.depra.ai/mcp --header "Authorization: Bearer $DEPRA_TOKEN". The MCP setup guide shows how to set the variable on Windows.
- Step 4. Type /mcp in Claude Code and check that both servers are listed.
- Step 5. Ask a first question: "List my Depra AI projects and name my connected Shopify store. Do not change anything."
- Other apps. The MCP setup guide covers Cursor, VS Code, Claude Desktop, Codex and other MCP apps. Connectors work in Claude Desktop too (Claude Help Center, checked 7 Oct 2026). In Cursor, store data comes through Shopify's CLI connector.
Which checks can you run once both are connected?
Five recipes follow. Each is a prompt to try, with the Depra AI tools and Shopify action it should use. All five only read, and none was run end to end. What comes back depends on your own data.
In Depra AI itself, a prompt means one tracked buying question. The Shopify action names come from Shopify's help page, which also advises one task per request (Shopify Help Center, checked 7 Oct 2026). So if a prompt stalls, split it: the Depra AI part first, then the Shopify part. The MCP recipes page has more prompts, and the MCP tools reference describes every tool.
1. Compare the products ChatGPT shows with your catalogue
Depra AI tools: list_projects and get_shopping_summary. Shopify action: search_products.
- Step 1. Try: "Use the Depra AI tools to get the shopping summary for my project for the last 30 days. List the top products ChatGPT showed and the top merchants. Then use the Shopify connector to search my catalogue, and tell me which of those products are mine. Do not change anything."
- Step 2. Check three rows by hand. The assistant matches by name, and names can differ.
- Step 3. Top merchants are the stores the product cards point to.
2. Find buying questions where none of your products appear
Depra AI tools: get_shopping_summary, list_chats with the shopping filter, and get_chat. Shopify action: search_products.
- Step 1. Try: "Use the Depra AI tools to list the prompts that triggered products in the last 30 days. Open their shopping answers and list the prompts where no product shown is from my brand. For each, use the Shopify connector to name the product in my catalogue that fits best, or say I have none. Do not change anything."
- Step 2. Confirm two of those answers in Depra AI.
- Step 3. If you sell a product that fits and ChatGPT did not show it, that is a gap to look into. See product visibility.
3. Write a weekly summary of what changed
Depra AI tools: get_recent_changes, get_brand_report and get_shopping_summary. Shopify action: get-inventory-levels.
- Step 1. Each Monday, try: "Use the Depra AI tools to summarise this week for my project. Include the recent changes, my visibility for the last 7 days next to the previous 7, and my products that ChatGPT showed. Then use the Shopify connector to flag any of those products that are out of stock. Give the number of answers behind each figure. Do not change anything."
- Step 2. Read it before you share it. Check the number of answers behind each figure, because a figure from a few answers can mislead.
- Step 3. Stock figures come from Shopify, not from Depra AI. Depra AI sends no alerts, so you run this prompt yourself. More in SEO automation with AI agents.
4. List the sources that cite rivals and skip you
Depra AI tools: get_source_gaps and get_url_report. Shopify action: get-product.
- Step 1. Try: "Use the Depra AI tools to list the top 10 source gaps for my project. Skip rows marked low confidence and keep the ones marked recommendable. Show the cited pages for the top three. Then draft a short pitch to each site for one of my products, using only facts the Shopify connector returns. Do not send anything, and do not change anything."
- Step 2. The gap score is a rank from 0 to 100 inside your project, not a percentage.
- Step 3. Check each pitch, then send it yourself. See citation gap analysis.
5. Check your product titles against the searches ChatGPT runs
Depra AI tool: list_search_queries, the web searches ChatGPT ran behind its answers. Shopify actions: search_products and get-product.
- Step 1. Try: "Use the Depra AI tools to list 25 web searches ChatGPT ran behind answers in my project in the last 30 days. Give the number of answers that ran each. Then use the Shopify connector to read the titles and tags of my 10 closest products. Show the words in those searches that my titles and tags do not use. Do not update any product."
- Step 2. Treat the result as ideas. No one can promise that a new title changes an answer.
- Step 3. Make the edits you agree with in your Shopify admin.
What are the limits of this setup?
The limits come from what each server reads and where it runs.
- Depra AI only reads. Its MCP server reads the stored data of a project. It cannot change a project, start a scan, or edit a store or a website.
- Product cards come from ChatGPT only. Gemini, Perplexity and Google AI Overviews are tracked for brand mentions and sources. Shopping percentages are shares of shopping answers, not of all answers.
- Nothing from Depra AI is installed in Shopify. DepraBot, Depra AI's web reader, reads your store's product list so Depra AI knows which products are yours. A Depra AI app for Shopify is coming to the Shopify App Store. When it is listed, it will read your products, collections and store pages. Until then, setup needs no Shopify app and no access token. See how Depra AI reads a store catalogue and Depra AI for Shopify stores.
- Not every app can connect. The Depra AI MCP server works with Claude Code, Cursor, VS Code and Codex, and with Claude Desktop through the mcp-remote bridge. Other MCP apps follow the Other tab on the API Keys page. It cannot connect to claude.ai on the web or to ChatGPT yet.
How do you keep a store safe with an AI assistant connected?
Give the assistant the least access that does the job. Shopify's connector for Claude can change your store once you approve write access, so work through the first five points first. They draw on Shopify's help page (Shopify Help Center, checked 7 Oct 2026).
- Read the access level before you approve it. Your Shopify admin shows the data access the connector requests. Read-only access lets it view data. Write access lets it edit data and act for you. You can decline a request.
- Do not approve write access for a task that only reads. Every check in this guide only reads. If the tool later asks for more access, Shopify asks you to authorise it, and you can say no.
- Connect with a limited staff account. The tool can do no more than the Shopify permissions of the person using it.
- Know how to reduce access. You cannot lower an access level after you approve it. Uninstall the integration from your Shopify admin to end its access, then install it again with less.
- Know where the data goes. Once you authorise a tool, the data you selected leaves Shopify and falls under that provider's terms.
- Treat copied text as data. The Depra AI MCP server labels text copied from websites and AI answers with field names starting untrusted_. An assistant should read that text, not obey it.
- Say it in the prompt. End each prompt with "Do not change anything in my store." That is a request, not a lock: the access you approved is the real limit.