Your next shopper may not start on your homepage. They may start in ChatGPT, Microsoft Copilot, Google AI Mode, Gemini, or the Shop app.
Shopify Agentic Storefronts and MCP make it possible for these AI experiences to discover products, understand store data, and guide shoppers toward checkout.
This is a connective layer that lets AI agents search your catalog, build a cart, and hand a shopper straight to checkout.
I have spent the past few months testing this setup on live stores, and this guide covers exactly where things stand today.
What is a Shopify MCP server? A simple explanation
Model Context Protocol, or MCP, is an open standard from Anthropic that lets AI models talk to outside tools and data using one shared format.
In simple terms, UCP defines the shopping flow, and MCP gives AI agents a standard way to call those shopping tools.
Shopify uses UCP as the commerce rulebook and exposes those UCP-powered shopping capabilities through Shopify MCP servers. In simple terms, UCP defines the shopping flow, and MCP gives AI agents a standard way to call those shopping tools.
UCP is an open framework Shopify co-developed with Google, and it defines how an agent proves who it is, searches a catalog, builds a cart, checks out, and tracks an order, all through a consistent set of rules.
Shopify's MCP servers are the implementation of that framework.
Here is where a lot of confusion starts. There is no single "Shopify MCP." There are two different things people usually mean when they say “Shopify MCP.”
- Storefront MCP is for shopping agents. It helps AI agents search products, manage carts, answer policy questions, and move shoppers toward checkout.
- Shopify AI Toolkit / Dev MCP is for developers. It helps Claude Code, Cursor, Codex, VS Code, and similar tools work with Shopify docs, GraphQL schemas, Liquid validation, CLI workflows, apps, and themes.
How does the Shopify MCP server power AI agents?
Every eligible Shopify store gets a live Storefront MCP endpoint by default, reachable at https://{yourstore}.myshopify.com/api/mcp, with a second UCP-specific endpoint at /api/ucp/mcp for catalog search.
No developer, code, or apps are required here. This single server handles the entire buyer journey in four stages.
Stage 1: Negotiate and authenticate
Every agent presents a profile that tells Shopify who it is. Based on that profile, Shopify assigns a trust tier, and the tier decides how much an agent can do on its own.
Stage 2: Discover products
Agents search your catalog through search_catalog, lookup_catalog, and get_product. Shopify also runs a cross-store Global Catalog, so an agent can search millions of products from every Shopify merchant at once, then narrow down to yours.
Stage 3: Build a cart and check out
The get_cart and update_cart tools let an agent add items and fetch pricing. When the shopper is ready, create_checkout converts that cart into a real checkout session, calculating tax, shipping, and totals against your live store settings.
Stage 4: Monitor the order
After purchase, get_order and order webhooks let an agent or the shopper check fulfillment status, refunds, and shipping updates without emailing your support team.
For the vast majority of agents today, checkout does not complete inside the chat window. The agent receives a continue_url and hands the shopper to your actual Shopify checkout to finish paying.
If the shopper is redirected to your Shopify checkout, your normal checkout experience is mostly preserved.
But if the shopper uses an AI channel’s Shopify-powered direct checkout, some checkout customizations may not appear, especially upsells, custom fields, loyalty blocks, custom pixels, subscriptions, bundles, digital products, pickup, and local delivery.
Shopify storefront MCP vs the Shopify AI toolkit
Storefront MCP is consumer-facing. It is what powers a shopper's conversation with an AI agent that wants to buy something from your store. You do not build this. Shopify runs it for you.
The Shopify AI Toolkit is a developer-facing project. Shopify open-sourced it in April 2026, and it connects coding assistants like Claude Code, Cursor, and VS Code to Shopify's documentation, GraphQL schema, and CLI so a developer can build themes, apps, and integrations faster.
The Toolkit's Dev MCP server gives an AI coding agent live access to Shopify's platform knowledge. This does not include your storefront's live sales data.
If someone asks you about MCP for Shopify and mentions building an app or theme, they mean the AI Toolkit.
If they mean letting ChatGPT or Claude sell your products to a shopper, they mean Storefront MCP.
Shopify's own AI assistant lives inside your admin and helps you run the store rather than helping outside agents sell from it.
How to get started with Shopify MCP: For store owners
Go to Sales channels > Agentic in your Shopify admin. From there you can:
- See which AI channels have access to your products through Shopify Catalog. This can be ChatGPT, Microsoft Copilot, Google AI Mode, Gemini, and Shopify's own Shop app.
- Turn off Allow Shopify to manage for me if you want manual control instead of automatic enrollment in new AI channels.
- Deactivate direct checkout on channels that support it, so shoppers are always redirected to your own checkout instead.
- Preview how your products actually appear in AI catalog search results.
A few requirements gate eligibility:
- You need your Terms of Service, Privacy Policy, and Return and Refund Policy filled out under Settings > Policies,
- Your products need to sell to US buyers,
- You need to accept Shopify's Agentic Storefronts Supplemental Terms of Service.
B2B-only products and anything on OpenAI's prohibited list will not surface in ChatGPT results.
An agent can only sell what it can understand, so titles, descriptions, variant details, and policy answers need to be complete and specific.
This overlaps heavily with answer engine optimization work, and I have a full breakdown of that in my Shopify AEO and GEO guide if you want to tighten up how your store reads to any AI system.
Claude Code Shopify MCP: Setup for developers and agencies
If you are building something custom, testing the protocol directly, or wiring Claude into your day-to-day store management, this is the section for you.
For coding work, the recommended path is the Shopify AI Toolkit plugin. Inside a Claude Code session.
- claude plugin install shopify-ai-toolkit@claude-plugins-official
- claude mcp add --transport stdio shopify-dev-mcp -- npx -y @shopify/dev-mcp@latest
Restart Claude Code, and the plugin gives your session access to Shopify's live GraphQL schema, Liquid validation, and CLI commands.
You need Node.js 18 or higher. The Dev MCP server does not require authentication because it connects to Shopify developer resources such as docs and schemas.
However, if you use Shopify CLI or store-execution capabilities to work with a real store, you should expect authentication and permissions to matter.
I walked through that exact process end to end, screenshots included, in my Claude and Shopify connection guide. This includes how to scope the token to read-only access so Claude can answer questions without being able to change anything until you are ready to grant write access.
What happens at checkout when an AI agent buys from you
Merchants ask me this more than anything else.
In almost every real case today, the buyer finishes payment on your Shopify checkout, either inside an in-app browser (ChatGPT) or a new browser tab.
You remain the merchant of record. The order lands in your admin with channel attribution showing it came from an AI referral, same as any other sales channel.
But some checkout blocks, particularly custom fields, upsell prompts, or specific validation rules, may not render inside an agentic storefront's direct checkout the same way they do on your normal storefront checkout.
The purchase still completes correctly, but any app-based messaging tied to those blocks might not show up.
If you rely heavily on checkout customization for revenue, it is worth auditing this now.
To wrap it up, here’s a checklist to get started with Shopify MCP
| Area | What to check |
| Agentic settings | Go to Sales channels > Agentic and review which channels have Shopify Catalog access. |
| Product data | Make titles, descriptions, variants, images, price, and availability complete. Shopify says these fields are structured for AI channels through Shopify Catalog. |
| Policies | Add Terms of Service, Privacy Policy, and Return/Refund Policy. These are required for ChatGPT and direct checkout eligibility. |
| US selling | Check whether the store sells to US customers. ChatGPT and Copilot require selling to US buyers; Google/Gemini direct checkout is more limited |
| Checkout customizations | Audit checkout apps, upsells, custom fields, validation rules, local delivery, pickup, subscriptions, bundles, and custom pixels. |
| B2B products | Make sure B2B-only products are not accidentally exposed if using custom B2B logic. Shopify says agentic storefronts support D2C sales only. |
| AI discoverability | Use Shopify’s Agentic preview/search tool as a directional signal, not a guaranteed ranking result. |
FAQs
1. What is Shopify MCP?
Shopify MCP refers to Shopify's Model Context Protocol servers. It is a layer that lets AI agents like Claude, ChatGPT, and Copilot search a store's product catalog, manage a cart, and complete a checkout using natural language.
2. How to get started with Shopify MCP?
Most merchants do not need to build anything. Check Sales channels > Agentic in your Shopify admin to see which AI platforms already have access to your catalog, confirm your store policies are complete, and review your product descriptions so agents have accurate information to work with.
3. What is the difference between Shopify Storefront MCP and the Shopify AI Toolkit?
Storefront MCP is the consumer-facing server that lets AI shopping agents browse and buy from your live store. The Shopify AI Toolkit is a separate, developer-facing tool that connects coding assistants to Shopify's documentation and schema so developers can build apps and themes faster.
4. How does Claude Code connect to Shopify MCP?
For development work, install the Shopify AI Toolkit plugin inside Claude Code with /plugin marketplace add Shopify/shopify-ai-toolkit followed by /plugin install shopify-plugin@shopify-ai-toolkit. Connecting Claude to your live store data for admin tasks uses a separate Admin API token setup instead.
5. Do AI agents complete checkout without a shopper ever visiting my store?
In most current cases, no. Shoppers are handed a link back to your own Shopify checkout to finish payment, so your branding and payment methods stay intact. Only a small number of vetted, high-trust agent partners can complete a purchase directly inside their own interface.
6. Can I stop AI platforms from selling my products?
You can turn off Shopify Catalog access per channel from Sales channels > Agentic and disable direct checkout where it is offered. Fully hiding a product from all AI discovery requires Unlisted status or the seo.hidden metafield, both of which also remove it from regular search engines, so it is rarely the right tool for a partial opt-out.
About the author
Vineet Nair
Vineet is an experienced content strategist with expertise in the ecommerce domain and a keen interest in Shopify. He aims to help Shopify merchants thrive in this competitive environment with technical solutions and thoughtfully structured content.