What is the Shopify AI Toolkit?

The Shopify AI Toolkit is an official open-source plugin and MCP server from Shopify that connects AI coding tools, Claude Code, Cursor, OpenAI Codex, Gemini CLI, and VS Code, to the Shopify developer platform. It ships agent skills, a Dev MCP server, and access to Shopify CLI's store execute capabilities, so the agent can search Shopify docs, validate GraphQL/Liquid/UI extension code, and run store operations without copy-pasting between the editor and the browser.

Is the Shopify AI Toolkit for merchants or developers?

Developers. It is an editor and CLI integration aimed at engineers building Shopify apps, themes, custom storefronts, and feed/catalog automation. Merchants who do not write code use Shopify Magic and Sidekick inside admin instead. The two stacks are complementary but separate.

What agent skills ship in the Shopify AI Toolkit?

Twenty skills as of release: shopify-admin (Admin GraphQL), shopify-storefront-graphql, shopify-custom-data (metafields and metaobjects), shopify-customer (Customer Account API), shopify-functions, shopify-hydrogen, shopify-liquid, shopify-partner, shopify-payments-apps, shopify-pos-ui, four Polaris extension skills, shopify-use-shopify-cli, shopify-app-store-review, shopify-dev (general docs), shopify-onboarding-dev, shopify-onboarding-merchant, and a ucp skill for the Universal Commerce Protocol.

How does the Shopify AI Toolkit relate to MCP?

The toolkit includes Shopify's Dev MCP server, which exposes Shopify's documentation, GraphQL schemas, and validators as MCP tools. Any MCP-compatible agent can connect, not just the five officially supported editors. The MCP server is local, runs without authentication for docs and validation, and pairs with the Shopify CLI for authenticated store operations.

Can the Shopify AI Toolkit edit a live Shopify store?

Yes, through Shopify CLI's store execute capabilities. The agent generates commands that the CLI runs against an authenticated store. This includes catalog operations like adding products, updating variants, managing metafields, and configuring channels, useful for catalog and feed automation work where the agent needs to make real changes rather than only suggest code.

Shopify AI Toolkit: What the Dev Plugin Unlocks for Feed and Catalog Automation

The Shopify AI Toolkit is the official plugin Shopify ships to connect AI coding tools to the Shopify developer platform. It is not Magic. It is not Sidekick. It is the developer-facing piece: an agent plugin that gives Claude Code, Cursor, OpenAI Codex, Gemini CLI, and VS Code direct access to Shopify’s documentation, API schemas, code validators, and store execute capabilities through the Shopify CLI.

For anyone writing catalog automation, feed exporters, metafield migrators, custom channel integrations, or agent-driven store operations, the toolkit changes the loop. This guide covers what it actually ships, how it works, and where it earns its keep on feed and catalog work.

What ships in the Shopify AI Toolkit

The toolkit is three layers stacked into one plugin.

Agent skills

Twenty skills as of the current release, each scoped to a specific Shopify surface. The agent loads them lazily, so it pulls the right docs and conventions in only when the task touches that surface.

shopify-admin, Admin GraphQL queries and mutations
shopify-storefront-graphql, Storefront API queries for custom storefronts
shopify-custom-data, metafields and metaobjects
shopify-customer, Customer Account API
shopify-functions, Shopify Functions (Discount, Cart Transform, Delivery Customization, etc.)
shopify-hydrogen, Hydrogen storefront recipes
shopify-liquid, theme templates, sections, blocks, snippets
shopify-partner, Partner API for app/affiliate data
shopify-payments-apps, Payments Apps API
shopify-pos-ui, POS UI Extensions
shopify-polaris-app-home, embedded admin app UI
shopify-polaris-admin-extensions, Admin UI Extensions
shopify-polaris-checkout-extensions, Checkout UI Extensions
shopify-polaris-customer-account-extensions, Customer Account UI Extensions
shopify-use-shopify-cli, CLI commands and store execute workflows
shopify-app-store-review, App Store submission compliance checks
shopify-dev, general developer docs search
shopify-onboarding-dev, getting started building on Shopify
shopify-onboarding-merchant, store setup and migration workflows
ucp, Universal Commerce Protocol

Dev MCP server

The plugin includes Shopify’s Dev MCP server, which exposes the same docs, schemas, and validators as MCP tools. Any MCP-compatible agent can connect, the five officially supported editors are not a hard ceiling. The MCP server runs locally with no authentication required for docs search and code validation.

Store execute through Shopify CLI

The agent can ask the CLI to run real store operations: query the Admin API, run mutations against products/variants/metafields, configure channels, manage themes. Authentication is the same shopify auth flow you already use. The agent never holds credentials; it generates commands the CLI executes against your authenticated session.

How the loop changes for feed and catalog work

The interesting question for feed-and-catalog work is not “does it write Shopify code.” Coding agents already do that. The interesting question is what the toolkit eliminates from the loop.

No more guessing at Admin schema

The shopify-admin skill plus the validate_graphql validator means the agent stops inventing fields. Catalog operations against Shopify’s product graph are heavy on mutations: productCreate, productVariantsBulkUpdate, metafieldsSet, inventorySetQuantities. Each has version-specific argument shapes the agent reliably gets wrong without a live schema. The toolkit fixes that.

Metafield and metaobject schemas in context

shopify-custom-data is the unsung hero for feed work. Metafields under the google namespace (material, pattern, age_group, gender, product_highlight, product_detail) are the soft-attribute backbone of any Merchant Center feed. The skill keeps the agent honest about definition shape, ownership, and value types. For metaobject-backed feeds, increasingly the norm for richly modelled catalogs, it is the difference between agents that can model the data and agents that can not.

For a deeper read on how those metafields map to Merchant Center, see how AI Max reads your Shopify product catalog.

Store execute closes the loop

The biggest practical shift. Before: agent generates a mutation, you copy it, you paste it into GraphiQL or your CLI, you run it, you paste results back. After: agent generates the mutation, the CLI executes it under your authenticated session, the result lands back in the chat.

For catalog cleanup work, bulk metafield population, GTIN backfill, vendor field correction, taxonomy mapping, this collapses a 20-step workflow into a single agent turn.

Validator-first iteration

The validators (validate_graphql, validate_components, validate_functions, validate_theme) emit a consistent markdown summary with an artifact ID and revision number. The agent chains retries cleanly: validate, fix, re-validate, all without leaving the editor. For Functions work and theme work especially this turns “compile errors at deploy time” into “errors caught before the file is written.”

The UCP skill matters for feed strategy

The ucp skill is worth calling out separately. Universal Commerce Protocol is the joint Shopify-and-Google standard for agentic commerce, the protocol that lets AI shopping agents discover and transact against any merchant. Shopify is shipping a first-party skill for it.

The practical read: feed strategy in 2026 is no longer just feeds-as-files. It is feeds-as-API-endpoints (Shopify Catalog API, MCP) plus UCP-discoverable storefronts. The AI Toolkit ships the developer surface for both. For the merchant-side checklist on UCP, see the universal commerce protocol merchant checklist.

Where the toolkit stops

Worth being explicit about what the toolkit does not do, because that is where you still need the rest of the stack.

It does not push to Merchant Center

The toolkit operates on the Shopify side. Merchant Center, Meta Catalog, TikTok Shop, Amazon, the long tail of marketplaces, those are separate APIs. You can still drive them from the same agent session through their own MCP servers (Google Shopping MCP, Meta Ads MCP, TikTok Ads MCP). See MCP vs REST API vs OpenClaw for Shopify product feed automation for the cross-platform pattern.

It does not handle multi-region or multi-currency feeds

The shopify-admin skill knows the schema. The orchestration of per-region price, availability, currency, and tax across Shopify Markets and downstream channels is still operational logic you write. See Shopify Markets product feed management.

It does not do bulk disapproval remediation

The agent can fix one product at a time using productUpdate or metafieldsSet. For thousands of disapprovals across categories, the better pattern is a dedicated feed app or a batch script against the Content API. The toolkit accelerates writing that script; it is not the script.

Telemetry is on by default

Both the docs search and the validators send usage events to shopify.dev/mcp/usage. The payload includes query text, validated code, validation results, file lists, and theme paths. For client work or sensitive catalogs, set OPT_OUT_INSTRUMENTATION=true before installing. Make the decision once, bake it into shell config.

Setup in three steps

A clean install for Claude Code, which is the most common starting point.

In a Claude Code chat run /plugin marketplace add Shopify/shopify-ai-toolkit then /plugin install shopify-plugin@shopify-ai-toolkit. The plugin appears in your plugin list and auto-updates.
Run shopify auth in your terminal against the store you want the agent to operate on. This is the same auth used for normal CLI work. Store execute then routes through this authenticated session.
Decide your telemetry posture. If you want the data shared, leave the default. Otherwise export OPT_OUT_INSTRUMENTATION=true in your shell config so it applies to every session.

For Cursor, install from the Cursor Marketplace. For Gemini CLI, run gemini extensions install https://github.com/Shopify/shopify-ai-toolkit. For OpenAI Codex, use /plugins and search for Shopify. For VS Code, use the Command Palette and Chat: Install Plugin From Source with the GitHub URL.

When the toolkit replaces what you were building

A common pattern in 2025 was building a custom MCP server that wrapped the Shopify Admin API so agents could drive a store. The toolkit makes that work obsolete for the official surfaces. If you have a homegrown shim around productCreate and metafieldsSet, it is worth deleting and switching to the toolkit’s surface, fewer moving parts, automatic updates, and the validators come for free.

The work that still belongs in custom MCP territory: channel-specific operations (Merchant Center, Meta Catalog, TikTok Shop, Amazon), proprietary business logic, multi-tenant agency tooling, and anything that needs to bridge Shopify with destinations the toolkit does not cover.

Where to go next

For the cross-platform agent-tooling pattern see MCP vs REST API vs OpenClaw for Shopify product feed automation. For the Merchant Center side of the same loop see Google Shopping MCP for Shopify stores. For the AI Max story on the Shopify catalog see how AI Max reads your Shopify product catalog.

The Shopify AI Toolkit is the developer-side counterpart to the merchant-facing AI work Shopify has been shipping. For teams automating Shopify catalogs, feed exports, and agentic commerce surfaces, it removes a meaningful amount of the boilerplate from the loop, and points the same agent at the official Shopify surfaces by default. That is a quietly significant shift for anyone whose job is keeping a Shopify catalog clean, optimised, and discoverable across every channel that matters.