Shopify Storefront MCP: A Merchant Guide to AI Shopping

Shopify Storefront MCP gives AI applications a standard way to search a merchant's catalog, retrieve product details, answer policy questions, and support shopping actions with current Shopify data.

The short answer: Storefront MCP is not a chatbot by itself. It is the commerce connection behind a chatbot, shopping assistant, or external AI agent. Shopify hosts the product, policy, and commerce tools; an AI application calls those tools, decides what to say, and presents the experience to the shopper.

For merchants, the most important work is not choosing a language model. It is making sure products, variants, attributes, availability, markets, and policies are complete enough for an agent to represent the store accurately.

MCP and UCP in Plain English

Two acronyms appear throughout Shopify's agentic commerce documentation.

MCP: Model Context Protocol

MCP standardizes how an AI application discovers and calls tools. Instead of writing a different proprietary connector for every model, a developer can connect an MCP-capable agent to tools with defined names, inputs, and outputs.

In this case, the tools expose Shopify commerce capabilities.

UCP: Universal Commerce Protocol

UCP standardizes commerce actions across the buyer journey: product discovery, carts, checkout, and orders.

Shopify's current Catalog MCP tools implement UCP and use the MCP transport. A useful mental model is:

MCP describes how an agent communicates with tools.
UCP describes the commerce capabilities and data shapes those tools use.

The protocols solve different parts of the same integration.

What Storefront MCP Can Do

Shopify documents tools across several commerce stages.

Search a store's catalog

search_catalog accepts a natural-language query such as:

A lightweight waterproof jacket for daily cycling in London under £180.

The response can contain matching products, price ranges, media, and variants from the selected merchant.

This is more useful than a basic keyword match when the catalog contains the attributes needed to understand "lightweight," "waterproof," "cycling," and the shopper's market.

Look up a known product

lookup_catalog retrieves products or variants when the agent already has an identifier from a prior search, saved item, or shared link.

Use this to refresh product information instead of trusting a stale answer from conversation history.

Resolve options and variants

get_product returns detailed product information and supports option selection. An agent can help a shopper narrow a product to a valid color, size, or other option while checking whether combinations exist and are available.

Answer policy questions

search_shop_policies_and_faqs can answer questions such as:

Can I return a sale item?
How long does delivery take?
Do you ship to Singapore?
Is this product covered by a warranty?

The quality of the answer depends on the merchant's published policy and knowledge content.

Work with carts and checkout

Shopify's agentic commerce documentation covers cart construction, localization, checkout handoff, and—at higher trust levels—direct checkout capabilities.

The exact endpoint and tool set is evolving as Shopify moves commerce capabilities toward UCP. Developers should inspect the live server tools and current UCP documentation instead of hardcoding assumptions from an older tutorial.

Support authenticated customers

Shopify also provides a Customer Accounts MCP server for customer-specific actions such as order lookup and account information.

This is a separate, authenticated integration. It requires OAuth 2.0, appropriate access scopes, and Level 2 protected customer data access. A public product assistant does not automatically have permission to read orders or personal information.

The Buyer Journey

A typical store-specific shopping conversation works like this:

A shopper describes what they need.
The AI application calls Storefront Catalog MCP.
Shopify returns matching product data.
The agent explains suitable options and asks useful follow-up questions.
The shopper chooses a product and variant.
The application creates or updates a cart.
The shopper is handed to Shopify checkout, or an approved trusted agent continues through supported checkout capabilities.

The AI application controls the conversation. Shopify remains the source for products, variants, prices, availability, carts, and checkout.

Storefront Catalog vs Global Catalog

Shopify provides two UCP catalog interfaces.

Storefront Catalog

Storefront Catalog searches within one merchant's store.

Use it when:

The shopper is already on a brand's storefront
A brand wants its own AI shopping assistant
The conversation should only recommend that merchant's products
Product comparisons happen within one catalog

The documented endpoint is:

https://{storeDomain}/api/ucp/mcp

Global Catalog

Global Catalog searches across Shopify merchants.

Use it when:

An agent helps a shopper compare products across brands
The shopper has not selected a merchant
Discovery begins with a general product need

The documented endpoint is:

https://catalog.shopify.com/api/ucp/mcp

Both catalog interfaces expose search_catalog, lookup_catalog, and get_product, but their scope and supported extensions differ.

What Is an Agent Profile?

UCP catalog requests include a URL pointing to the agent's profile.

The profile tells Shopify:

Which agent is making the request
Which capabilities the agent supports
How Shopify can verify it
Which trust tier, rate limits, and tools may apply

An agent profile is not a merchant API password. Shopify's catalog documentation describes profile-based access without an API key for standard access, while higher access and rate limits can require additional approval or authentication.

Developers should treat the profile as part of the agent's identity and capability negotiation, not as a value to copy from an example indefinitely.

What Merchants Get and What Still Needs Building

It helps to separate Shopify infrastructure from the customer experience.

Shopify provides

Store and global catalog interfaces
Structured product and variant responses
Catalog search and product lookup tools
Cart, checkout, and order capabilities documented through UCP
Store policy and FAQ retrieval
Customer account tools with protected access
Live commerce state from Shopify

The merchant or development team provides

The customer-facing chat or agent interface
Model selection and AI orchestration
Conversation history and application state
Guardrails and response rules
Branding, accessibility, and analytics
Escalation to human support
Testing against the merchant's real catalog
Any custom tools outside Shopify's standard capabilities

Storefront MCP does not place a finished AI assistant on the storefront by itself.

Why Product Data Quality Determines Recommendation Quality

An agent cannot reliably infer a fact the catalog never states.

Consider a shopper asking for:

A fragrance-free moisturizer for sensitive skin that is vegan and safe during pregnancy.

If the product title says only "Daily Cream" and those attributes are absent or inconsistently buried in prose, the agent has weak evidence for a recommendation.

Strong catalog data includes:

Clear product titles
Specific descriptions
Shopify taxonomy categories
Consistent option names
Unique SKUs and valid identifiers
Materials or ingredients
Dimensions, fit, and sizing
Compatibility
Certifications
Intended use
Care instructions
Accurate prices and availability
Useful product media

Store stable facts in consistent product fields and metafields. Do not create a different field name for the same attribute in each collection.

Variants Need More Than a Color Name

Variant quality affects both recommendations and cart accuracy.

Audit:

Whether every valid option combination exists
Whether unavailable and nonexistent combinations are distinguished
Whether SKUs are unique
Whether prices are correct by variant
Whether variant images show the selected option
Whether size labels are consistent
Whether market availability is accurate

An agent may find the right parent product and still fail the shopper if it cannot resolve a purchasable variant.

Policies Must Be Written for Retrieval

Policy tools work best when the source material contains direct, unambiguous answers.

Weak:

We want every customer to love their purchase. Reach out and our team will see what it can do.

Stronger:

Unworn full-price items can be returned within 30 days of delivery. Sale items are final sale. Customers in the United States receive a prepaid return label.

The stronger version tells a shopper:

Which products qualify
The return window
The start date for that window
Whether sale items differ
Who pays return shipping
Which market the policy applies to

Review shipping, returns, warranties, subscriptions, exchanges, and product-specific exceptions. Keep storefront copy, Shopify settings, structured data, and support answers consistent.

Markets and Localization

An AI shopping experience should not recommend a product based on the wrong market.

Pass relevant buyer context where supported and verify:

Currency
Market price
Product availability
Shipping destination
Language
Duties and tax presentation
Variant availability
Checkout localization

Never let a conversational answer override Shopify's current commerce state. The cart and checkout response should remain authoritative for final totals and availability.

Storefront MCP Is Not the Same as Product Schema

Product JSON-LD and Storefront MCP serve related but different channels.

Product structured data helps web crawlers understand a product page.
Merchant Center feeds help Google verify and distribute product information.
Shopify Catalog and MCP tools expose commerce data to agents.

A merchant should maintain all three where relevant. They should agree on product identity, price, availability, brand, images, and variants.

Our Product Schema for Shopify Hydrogen guide covers the web structured-data layer. The Hydrogen AI search audit covers rendering, crawlability, feeds, and technical consistency.

Does a Merchant Need a Custom AI Assistant?

Not every store needs one.

A custom assistant can make sense when

The catalog is large or technically complex
Customers need help comparing specifications
Compatibility determines whether a product works
Product discovery normally requires several filters
The store has detailed policies or configuration steps
Bundles or routines require guided selection
B2B buyers repeat complex orders
Support receives the same pre-purchase questions every day

A custom assistant may be premature when

The catalog has only a few straightforward products
Product data is incomplete
Policies are unclear
The team cannot monitor incorrect answers
Existing navigation and search have obvious unresolved problems
The assistant has no job beyond repeating product descriptions

Fix the underlying catalog and storefront first. A conversational interface can make good data easier to use; it cannot turn missing data into trustworthy advice.

A Sensible Implementation Architecture

A production storefront assistant usually needs:

Chat interface — embedded storefront UI with accessible controls and clear loading states.
Application server — protects model credentials, manages sessions, and enforces rules.
AI model — understands the shopper's request and decides when to call a tool.
Shopify MCP client — discovers and invokes catalog, policy, cart, and account tools.
Conversation store — retains only the context required for a coherent session.
Analytics — records tool success, failed searches, handoffs, and conversions without exposing sensitive prompts.
Human escalation — transfers questions the assistant cannot answer safely.

Do not call paid model APIs directly from storefront JavaScript with a secret key.

Guardrails That Matter

An assistant should:

Use current tool results for price and availability
Say when information is missing
Avoid inventing materials, compatibility, delivery dates, or policy exceptions
Confirm a variant before adding it to cart
Show the shopper what changed in the cart
Require authentication before accessing customer data
Avoid exposing private conversation or account information in analytics
Offer a human handoff for uncertain or sensitive requests

The most dangerous answer is not an obvious error. It is a confident answer built from incomplete catalog data.

How to Measure Value

Track outcomes, not conversation volume.

Useful measures include:

Product searches that return no useful result
Product-detail tool failures
Variant-selection completion
Add-to-cart rate after an assisted conversation
Checkout handoff rate
Assisted conversion rate
Support handoff rate
Repeated unanswered questions
Revenue by assisted session
Return or cancellation reasons associated with incorrect recommendations

Review conversation samples with privacy controls. Failed queries often reveal missing product attributes or policy gaps that should be fixed at the source.

Merchant Readiness Checklist

Questions to Ask an Implementation Partner

Which Shopify MCP and UCP capabilities will the assistant use?
How will the integration adapt when live tool schemas change?
Which product attributes are missing today?
How are markets and buyer context handled?
Where is conversation data stored, and for how long?
How are model and tool errors shown to shoppers?
How does a shopper correct or remove a cart item?
When does the assistant require authentication?
What happens when the model is uncertain?
Which metrics prove the assistant is helping?

An implementation proposal should answer these questions before discussing animation, avatars, or the personality of the chat bubble.

Need Help Planning a Shopify AI Shopping Experience?

We build custom Shopify apps, Shopify Hydrogen storefronts, and the product-data foundations they depend on.

If your catalog is ready but product discovery remains difficult, contact Webmakers Studio. We can map the customer questions, Shopify data, MCP tools, guardrails, and measurement plan before deciding whether a custom assistant is justified.