Use cases by system9 use cases

What you can build on Shopify with AI, past Sidekick and the app store

Shopify already ships more AI than most platforms. Sidekick works for the merchant inside the admin: it analyzes sales, edits products, manages orders, runs background tasks and can generate small custom apps. On the other side, Shopify's agent layer, built on the Universal Commerce Protocol, gives shoppers' AI assistants catalog, cart, checkout and order tools, and since the Spring '26 Edition catalog access takes an API key with no approval.

Neither one talks to your customers about their order, and neither reads the systems Shopify does not hold. The builds in this hub sit in those joins: order status answers when the real ship date is in the 3PL, supplier ETAs that hide in email until a best-seller runs out, a Monday brief that explains last week instead of charting it, and an AI-readable catalog so assistants describe your products correctly.

Shopify is a pleasant platform to build on: a GraphQL Admin API, webhooks for every order event, and Shopify Flow for simple rules. The constraints that shape a design are cost-based rate limits, webhook delivery and the protected customer data rules.

What to build around Shopify

What the Admin API allows, and the limits a build respects

  • Rate limits scale with the plan

    The GraphQL Admin API meters query cost: 100 points a second on standard plans, 200 on Advanced, 1,000 on Plus and 2,000 on Enterprise, with 1,000 points as the ceiling for a single query. Live lookups fetch one order by name; backfills run as bulk operations.

  • Webhooks are at-least-once

    An orders/updated event can arrive twice, late, or before orders/create. Each delivery is verified by its HMAC signature and deduplicated on the ID header Shopify sends with it, and a scheduled reconciliation query catches anything that never arrived.

  • Customer data needs a reason

    Names, emails and addresses fall under Shopify's protected customer data rules, so the app requests only the fields each step needs. An order status agent needs the order's email to verify the customer, not the customer's whole history.

  • Refunds and cancellations are code decisions

    A refundCreate or orderCancel mutation commits money. The agent can propose one, but thresholds, eligibility windows and fraud checks live in code, and anything above the limit goes to a person with the order open.

  • Flow before custom code

    Shopify Flow already tags risky orders, alerts on low stock and routes simple events at no extra cost. I build only what Flow cannot do: read an email, judge an exception, or call a system outside Shopify.

Where Sidekick, the agent catalog and helpdesk apps stop

  • Sidekick is for staff, not shoppers

    It answers the merchant's questions about the store and makes changes in the admin. It does not reply to customers, cannot see your 3PL or ERP, and is not a support channel.

  • The catalog serves other people's agents

    The Catalog, Cart, Checkout and Order MCP servers let shopping assistants find and buy your products. The work on your side is product data those agents read correctly: attributes, variants, availability and policies, not another chatbot.

  • One store, one source of truth: buy

    When every answer is in Shopify and the returns policy is standard, a helpdesk agent billed per resolution is the cheaper answer, and I say so before scoping anything.

  • Custom starts at the second system

    A 3PL like ShipBob, a regional carrier, an ERP holding stock and wholesale terms, several expansion stores, or WhatsApp as the main channel: each is a place the apps cannot read, and each is where a build pays.

Frequently asked questions

Can I build a custom AI agent for my Shopify store?

Yes. A custom agent runs as a Shopify app with only the scopes it needs, reads orders through the Admin API, listens to webhooks, and connects to the other systems your answers depend on: the 3PL, the ERP, the helpdesk, WhatsApp. It makes sense when the answer is not fully in Shopify. When it is, a helpdesk app is usually enough.

Is Shopify Sidekick an AI customer service agent?

No. Sidekick is an assistant for the merchant and staff inside the Shopify admin. It analyzes data, edits products, manages orders and can build small custom apps, but it does not reply to customers. Customer-facing support needs a helpdesk agent or a custom one, and either can run alongside Sidekick without conflict.

What is Shopify's Catalog MCP, and do I need it?

It is part of Shopify's agent layer on the Universal Commerce Protocol: endpoints that let AI shopping assistants search your catalog, build carts and check out. You do not build it; Shopify runs it. What you control is the product data behind it, so the useful project is making attributes, variants and policies complete, consistent and sourced from real specifications.

Will an AI integration hit Shopify's API rate limits?

Not if it is designed for them. Limits are measured in query cost per second and rise with the plan, from 100 points a second on standard plans to 2,000 on Enterprise. An agent answering questions makes small, targeted queries, and heavy work such as reindexing the catalog runs as bulk operations overnight. When Shopify throttles, the response says so and the client backs off.

Also browse: E-commerce · Zendesk and helpdesks · WhatsApp · Marketing and content · all use cases