Skip to content

Powered by

Shopify Premier Partner
Let's talk

Shopify MCP Explained: What It Exposes to AI Agents

3D illustration of a glowing central node connected by lines to six hexagonal nodes on a dark background

Summarize with AI

What Shopify MCP is and what it exposes to AI agents — the shopper side (UCP catalog, cart and checkout servers, Customer Accounts MCP, Shopify Catalog) and the builder side (Shopify AI Toolkit).

AI agents are starting to shop. A customer can ask ChatGPT, Gemini or Copilot to find a product, check whether it's in stock, and then buy it, either in your store's checkout or, on some channels, without leaving the assistant. The plumbing that makes this possible on Shopify is MCP, and it moves the first contact with your store into the assistant. Here's what it is and exactly what it exposes to AI agents, checked against Shopify's documentation as of November 2026. This area changes almost monthly, so we date the facts that move fastest.

This is a deep-dive companion to our overview, Shopify Agentic Commerce: AI as Shopper vs. AI as Builder. Start there for the full map; stay here for MCP in detail.

What is Shopify MCP?

MCP stands for Model Context Protocol — an open standard that lets AI applications connect to external systems through one consistent interface. The easiest analogy is a USB-C port: instead of a different cable for every device, you get one universal connector. MCP is that universal port between AI agents (ChatGPT, Claude, Gemini and others) and your store's data and actions.

What does Shopify MCP expose to AI agents?

At its core, Shopify's MCP exposes your store's live data and commerce actions:

  • Product catalogue — products, variants, descriptions, availability.
  • Cart state — what an agent has added, in real time.
  • Order information — when using an AI agent, logged-in customers can also perform tasks such as reordering and tracking orders.
  • Store policies — shipping, returns, and other policy content.
  • Checkout — completing the purchase through Shopify checkout, with your payment methods, shipping rules and discounts.

The catalogue, cart and checkout actions now run through Shopify's UCP endpoints rather than the original Storefront MCP tools. Shopify deprecated the old cart tools on 24 June 2026, kept them running until 31 August 2026, and has since removed the catalogue and cart tools from the old endpoint (changelog). The table further down shows which server does what.

In practice, this is what lets a shopping agent not just mention your products but take a customer through a full purchase: check stock in real time, build a cart, and hand over to checkout. Where that checkout happens depends on the channel. As of November 2026, Shopify's Help Center says ChatGPT sends shoppers to your online store checkout (in an in-app browser or a new tab), while Google AI Mode and Gemini, Microsoft Copilot and Meta surfaces can use Shopify-powered checkout inside the channel if you switch it on (Shopify Help Center).

The Ecosystem Architecture: How UCP and MCP Connect

To fully understand Shopify’s agentic strategy, it helps to separate the map from the plumbing.

  • Universal Commerce Protocol (UCP): the open standard Shopify co-developed with Google. The official site, ucp.dev, now lists Amazon, Etsy, Meta, Microsoft, Target, Walmart and Wayfair among its co-developers. UCP is the map: the shared language for how AI agents discover and buy from merchants. On Shopify you don't host a manifest yourself. Your products are discoverable through Shopify Catalog (Shopify's docs describe the Global Catalog as covering "hundreds of millions of Shopify listings"), and agents identify themselves with a UCP profile hosted at a well-known URL.
  • Model Context Protocol (MCP): once an agent has found products, MCP is one of the transports it uses to act on your store's live data (UCP also runs over REST, and supports A2A). UCP says, "Here is a store that sells shoes," and MCP lets the agent say, "Give me size 10 in blue, check inventory, and build a cart."

Beyond text: interactive components in the chat

Large language models work with text and JSON, and walls of text are a poor way to browse visual categories like fashion, home decor or beauty.

Shopify's engineering team described this problem in 2025 and adopted MCP UI, a way for an MCP server to return interactive components, not just text. The approach has since been folded into MCP Apps, the standard extension for interactive UI in MCP.

  • How it works: instead of the assistant typing out a product description, the server returns components such as product cards, which the assistant renders in a sandboxed frame.
  • Who stays in control: the components don't change the cart themselves. They send intents (for example "add to cart") that the agent acts on.

Why this matters: your product photography, variant selectors and prices can appear inside the conversation, rather than as a plain list of links.

LayerProtocol / TechnologyWhat it Does for the MerchantWhy it Matters
1. DiscoveryUCP(Universal Commerce Protocol)Makes your products findable by AI agents through Shopify Catalog.Agents read structured product data instead of scraping your pages.
2. Action & DataMCP(Model Context Protocol)Gives agents an endpoint on your store to read stock and build carts.One interface for every AI app, instead of a custom integration per assistant.
3. PresentationMCP UI / MCP AppsRenders product cards and size/colour selectors inside the chat.Your product photos and variants show up in the conversation, not just as text.

Two sides of Shopify MCP: AI as shopper, AI as builder

For a merchant, the useful split isn't by server name. It's by who benefits: your customers' AI assistants, or your own team's coding agents. The two sides have different owners, different budgets and different things to get ready.

1. The shopper side: demand that arrives through AI assistants. Every Shopify store exposes these servers, with no app to install. This is how Shopify's documentation splits the work as of November 2026:

What the agent doesServerTools (Shopify docs, Nov 2026)
finds productsStorefront Catalog MCP, via UCP (/api/ucp/mcp)search_catalog, lookup_catalog, get_product
builds a cartCart MCP, via UCPcreate_cart, get_cart, update_cart, cancel_cart
completes the purchaseCheckout MCP, via UCPauthenticated or signed requests only; trusted agents can complete checkout directly
answers policy and FAQ questionsAsk MCP, via UCPask_business (UCP draft spec; availability varies by store)
tracks ordersOrder MCP, via UCPget_order, plus webhooks for fulfilment, refunds and returns
helps logged-in customersCustomer Accounts MCPunchanged by the UCP move

Agents built on the old Storefront MCP catalogue and cart tools have to move to UCP (Shopify's migration guide). On the merchant side, the sales channel that puts your products into AI assistants is called Agentic Storefronts: it's active by default for eligible stores, and you choose the AI channels under Sales channels > Agentic in the Shopify admin.

Two newer pieces are worth knowing about. In October 2026 Shopify launched Shop MCP, one connection that lets any MCP-compatible agent search Shopify Catalog, build carts and complete Shop Pay checkout after the shopper confirms the total (changelog). And the knowledge base in the Agentic sales channel, where you publish facts about your store for AI agents, is replacing the Shopify Knowledge Base app, which Shopify deprecates on 15 January 2027 (Help Center).

Who owns it: e-commerce and merchandising, not developers. Agents read your product data, variants, stock, policy pages and FAQs. In our projects the work here is mostly data: complete attributes and variant names, policy pages that answer the questions shoppers actually ask (delivery times, returns, sizing), and stock that matches reality. A shopper asks an assistant what to buy, and the stores it suggests are the ones whose data a model can read without guessing.

2. The builder side: how fast your team ships. The Shopify AI Toolkit became generally available with the Spring '26 Edition in June 2026 (Shopify). It connects coding agents such as Claude Code, Codex, Cursor and VS Code to Shopify's docs and API schemas, validates GraphQL, Liquid and extension code, and can run supported store tasks through the Shopify CLI, with the user deciding when they run. It ships as a plugin, a single shopify agent skill, and the Dev MCP server (@shopify/dev-mcp), which runs locally (Shopify docs).

Who owns it: your development team or agency. We describe how we use it below.

The shopper side earns you AI demand; the builder side changes how quickly you can act on it. Plan them separately.

How we use Dev MCP and AI agents in Shopify builds

AI agents now do a large share of the implementation work in our builds. What makes that safe is not the tool but the process around it: the agents work inside our processes and guardrails, built on experience from our client projects, and a person from our team reviews and approves the work before it goes live. In practice:

  • People set the boundaries first. Our engineers design the architecture, the integrations with ERP, PIM or WMS, and the data model before an agent starts.
  • The agent works from Shopify's current docs, not from memory. We connect it to the AI Toolkit and Dev MCP so it checks APIs and schemas against live reference material and validates GraphQL before we see it. That cuts down on invented fields and outdated API calls, two mistakes we otherwise see often in AI-written Shopify code.
  • Guardrails limit what an agent can change on its own. Changes to a store or a backend system don't happen without a person's approval.
  • Nothing goes live unreviewed. Someone from our team checks the work, alongside automated tests, before it ships.

The full workflow, phase by phase, is in our AI-assisted Hydrogen workflow.

Why this matters now

This isn't a someday story. With the Spring '26 Edition (June 2026), Shopify removed the approval step for building on its agentic commerce layer: developers register an agent profile and call the public MCP endpoint. The standard underneath, UCP, was co-developed with Google and is backed by much of the industry, from Amazon, Meta and Microsoft to Stripe, Etsy, Target and Wayfair. It's shared infrastructure, not a Shopify-only experiment.

Shopify also says structured data wins: AI searches powered by Shopify Catalog convert at 2x the rate of those using scraped data (Shopify's own figure; it doesn't publish the underlying data). Whatever the exact multiple, the practical point stands: an agent can only recommend what it can read, so your product data, FAQs and policies need to be complete. (Source: Shopify, Spring '26 Edition, 17 June 2026.)

What AI agents can — and can't — do

Through MCP, agents can read and write your store data: query the catalogue, manage products, pull reports, manage carts and orders. You can even run parts of your store in natural language from an interface like Claude.

What they cannot do is touch Shopify's proprietary core. AI agents can read and write store data via MCP, but they can't read or modify Shopify's backend business logic — that lives on Shopify's servers. When you need truly custom backend behaviour, that's a development question, not an MCP one.

Do you need a headless store for MCP?

For AI shoppers — no. A Shopify store can be exposed to AI channels through MCP regardless of whether it's Liquid or headless (Hydrogen). Architecture is irrelevant for basic visibility in ChatGPT or Gemini.

Headless matters more for AI-assisted engineering — the developer workflow with Claude Code and Cursor works far better on a clean Hydrogen/TypeScript codebase than on Liquid. We've documented that workflow end to end.

MCP for B2B: connecting to your stack

B2B agentic commerce is where MCP gets powerful — and complex. It means connecting MCP to your operational systems: ERP for pricing and availability, PIM for product specs, WMS for inventory. The Shopify B2B + headless + MCP stack is feasible, but the value is in careful integration and middleware design.

Getting started (and doing it safely)

For AI shoppers, the switch is mostly on already: check Sales channels > Agentic in your admin, pick your channels and fill in the knowledge base. The work is in your data.

Letting agents act on your own store, from Claude or another assistant, is different. A production-ready setup involves MCP configuration, access controls and guardrail design: explicit limits on what an agent can read or change without human approval. This is where most of the real engineering work (and risk) lives.

If you'd rather listen, we cover the merchant side of agentic commerce in our podcast episode Agentic Commerce: 15x more orders from AI channels, and Marcin Kamiński explains why, in his measurements, AI bots visit stores far more often than people click through from AI answers, in E-commerce in the age of AI (both in Polish, with English subtitles and transcripts).

Want to make your store available to AI shoppers, safely? Talk to our Agentic Commerce team.

FAQ

What is Shopify MCP?

Shopify's implementation of the Model Context Protocol — an open standard (think USB-C for AI) that lets AI agents connect to your store's data and commerce actions through one interface.

What does Shopify MCP expose to AI agents?

Your live product catalogue, cart state, order information, store policies, and checkout: enough for an agent to find products and take a shopper through to Shopify checkout.

What are the Shopify MCP servers?

As of November 2026, two sides. Shopper side: Storefront Catalog, Cart, Checkout, Ask and Order MCP, all through the UCP endpoint (/api/ucp/mcp), plus Customer Accounts MCP for logged-in customers. Builder side: the Shopify AI Toolkit, including the Dev MCP server, for developers. Shop MCP (October 2026) is a separate single connection for agent builders who want Shopify Catalog search and Shop Pay checkout.

Do I need a headless store to use Shopify MCP?

No for AI-shopper visibility. Headless mainly helps the AI-assisted development workflow, not basic agent access.

Can I manage my Shopify store from Claude via MCP?

Yes, with setup: products, collections, content and reports through Claude, once MCP, access controls and guardrails are configured. We keep a person approving any change that writes to the store.

Author

Matt Czerniak

Co-founder at wecanfly | E-commerce & Shopify expert

Follow on LinkedIn

Let's start with a conversation

The ideal first step would be a short conversation during which you tell us about the challenge you are facing.

Black and white portrait of a young man with shoulder-length hair and a mustache, wearing a plaid shirt over a white t-shirt, against a background with plants and a grid wall.

Adam Choromański

New Business Manager