The four MCP layers of Shopify, MCP vs UCP, the June 15 deadline and how to get your store agent-ready.
Everyone in ecommerce is talking about MCP. Very few explain it clearly, and almost no one explains it well for Shopify. This is the guide we wish existed: every MCP layer Shopify shipped, what each one does, who it is for, and the one deadline you cannot miss. No hype, no jargon for its own sake.
MCP — the Model Context Protocol — is an open standard, created by Anthropic, that lets AI models connect to external tools and data in a consistent way. Anthropic released it as open source in late 2024, and every major AI platform has since adopted it.
Before MCP, every AI integration was custom: a different connector for each tool, each API, each bot. MCP replaces that with one protocol. Build the connection once, and any MCP-compatible AI — Claude, ChatGPT, Gemini, Cursor — can use it. For Shopify, that is the difference between an AI that can read your store and one that cannot.
The key mental model: MCP is a layer on top of your existing APIs, not a replacement for them. Your REST and GraphQL endpoints still exist. MCP makes them legible to AI agents.
The single most misunderstood fact about Shopify and MCP is that it is not one product — it is four distinct layers, shipped over six months. Merchants ask "should I turn it on"; developers ask "which endpoint". Both questions miss the structure. Here is the whole map.
Dev MCP is a local server that gives AI coding agents access to Shopify's developer documentation and GraphQL API schemas. It runs on your machine, needs no authentication for docs and schema access, and never touches production data. Its job: your AI coding assistant writes better Shopify code because it checks the live docs and schemas instead of relying on stale training data. It covers the Admin GraphQL API, Shopify Functions, Hydrogen and theme code.
Who needs it: developers and agencies. If you do not write code, you can skip this layer entirely.
Storefront MCP is the consumer-facing layer. Every Shopify store has a live MCP endpoint at /api/mcp that AI shopping assistants use to search the catalog, manage carts and answer policy questions on behalf of a shopper. It is what ChatGPT, Claude, Perplexity and Copilot talk to when a customer shops your store through them. Its core tools cover product discovery, cart management and store information (shipping, returns, FAQs).
Who needs it: every merchant — but there is nothing to install. The endpoint is automatic. Your job is the data behind it.
Catalog MCP is the global search layer. Where Storefront MCP queries one store, Catalog MCP lets an agent search across all eligible Shopify merchants in a single call. It requires authentication (JWT tokens, short-lived). This is what an agent queries when a shopper asks "find me running shoes under $120 that ship to Canada" — it hits Catalog MCP, not any individual store. Results cluster by a Universal Product ID so duplicate listings do not flood the agent.
Who needs it: agent builders and platforms. Most merchants never call it directly — but it is how AI assistants discover your products.
Customer Account MCP lets AI assistants help a logged-in shopper track orders, manage returns and access their account. Because it touches personal data, it uses an OAuth authentication flow. It is the layer that turns "where is my order" from a support ticket into a conversation.
Who needs it: merchants who want post-purchase support handled by AI. It is optional, and adds setup work because of the auth flow.
MCP and UCP are not competitors — they are complementary, and confusing them is the most common mistake in 2026 Shopify-AI discussions.
MCP (Model Context Protocol) is general-purpose: it defines how any AI model communicates with any data source. UCP (Universal Commerce Protocol) is commerce-specific: it defines the business logic of a shopping transaction — catalog structure, cart, checkout. Think of MCP as the transport and UCP as the commerce rulebook that travels over it. Shopify's Catalog tools are a UCP implementation delivered over MCP transport.
Why this matters to you: UCP is what makes your products comparable and buyable across every AI agent in a consistent way. It is the standard that turns "my catalog" into "a catalog an agent anywhere can transact against."
On April 22, 2026, Shopify moved the Storefront Catalog MCP server to the Universal Commerce Protocol. This is a breaking API change with a hard deadline. Here is the timeline that matters:
| Date | What happens |
|---|---|
| April 22, 2026 | UCP version announced; old endpoint and tool names deprecated |
| May 30, 2026 | New UCP endpoint and tool names become effective |
| June 15, 2026 | Old endpoint and old tool names are fully removed |
/api/mcp directly. You do NOT need to act if you run a standard Shopify or Hydrogen store with no custom agent tooling — the migration is handled for you. Unsure? Search your codebase for /api/mcp. No matches means nothing to do.The Shopify AI Toolkit — open-sourced under an MIT license on April 9, 2026 — is the recommended way to connect AI coding agents to Shopify. It bundles the Dev MCP server, agent skills and a plugin into one install. It needs Node.js 18 or higher and a supported tool: Claude Code, Codex, Cursor, Gemini CLI or VS Code.
The recommended route, in Claude Code, is the plugin — it updates automatically as Shopify ships new capabilities:
/plugin marketplace add Shopify/shopify-ai-toolkit
/plugin install shopify-plugin@shopify-pluginPrefer the MCP server directly? The manual route is one command. The server runs locally and needs no API key for documentation and schema access:
claude mcp add --transport stdio shopify-dev-mcp -- npx -y @shopify/dev-mcp@latestFor operations on a real store — editing themes, running bulk product changes — also install the Shopify CLI and authenticate. Then restart your AI tool to load the configuration.
MCP is model-agnostic: any AI model works through the MCP client layer. But in practice the three big assistants play different roles in the Shopify ecosystem. Tap each to compare:
| Shopping integration | Agent shopping live | Shopping mode live | AI Mode rolling out |
|---|---|---|---|
| Developer tooling | Claude Code + AI Toolkit plugin | Codex (skills + MCP) | Gemini CLI |
| Best for merchants | Store ops in plain English | Largest shopper reach | Google-surface discovery |
| Best for developers | Strongest agentic coding | Solid, IDE-friendly | CLI-first workflows |
| Note | Shopify Sidekick runs on Claude | Largest consumer install base | Tied to Google AI surfaces |
You do not install anything. Your win is discovery: when a shopper asks an AI assistant for a product you sell, Storefront and Catalog MCP decide whether you appear. The work is on the data side — clean titles, structured metafields, clear policy pages. A merchant's MCP strategy is really a product-data strategy. See our guide on connecting AI assistants to Shopify for the hands-on side.
Dev MCP and the AI Toolkit are your daily drivers. Your AI coding agent validates GraphQL against the live schema, edits themes, and runs bulk operations — tasks that used to mean custom scripts. The practical rule: docs and schema work first, store mutations only after testing on a dev store.
MCP changes delivery speed. An agency running the AI Toolkit across client projects iterates faster and audits stores at higher fidelity. The strategic shift: you stop being a store builder and become an infrastructure partner — the team that makes a client's catalog agent-ready before competitors realise the channel exists.
MCP exposes your store to AI agents — but agents only recommend stores whose data they can parse cleanly. Tick off every item below:
No. Storefront MCP is automatic — every store has a live /api/mcp endpoint. You only install something (the Dev MCP or AI Toolkit) if you are a developer building or operating stores.
Dev MCP is what your AI coding assistant uses to help you build. Storefront MCP is what other AI agents use to shop your store. Most operators only need the first.
The Dev MCP and the open-source AI Toolkit are free. Storefront and Catalog MCP carry no separate Shopify fee — they are part of the platform.
The old Storefront Catalog MCP endpoint and tool names are removed. Standard stores are unaffected. Anyone running a custom AI agent against /api/mcp must migrate to the UCP endpoint before that date.
All of them, automatically. MCP is a standard — optimising your product data exposes you to Claude, ChatGPT, Gemini and any other MCP-compatible agent at once. There is no per-assistant setup for merchants.
