Commerce
AI agent for Shopify: join store orders to ad spend
TL;DR
Shopify connects over a two-tier integration: Shopify's own Storefront MCP for catalogue and cart, plus admin reads for orders, products and inventory. The reason to connect it is not store reporting — Shopify does that well already — it is that your ad spend is not in the Shopify admin, so the question of whether a product earns more than it costs to sell needs both sides in one run.
Connecting it, and what it asks for
- 1Storefront-only needs nothing: give the agent your domain and it can read the public catalogue.
- 2For real sales data, open Integrations → Shopify → Connect store → Install from Shopify: Shopify asks you to choose the store and approve the app, then you connect it to a workspace — no store domain to type. The store's row shows which permissions Shopify granted.
- 3The Shopify app asks for product, inventory, discount and web-pixel write permissions, and not the permission to change orders. Every store change still pauses for approval.
The Shopify app's permissions, each with its reason, are listed in the Shopify guide. Every store change still pauses for approval, and through the app no order can be created, refunded, captured or cancelled. A first goal worth typing: Join last quarter of Shopify orders with ad spend and give me true ROAS by product.
What the agent does with Shopify
- Reads orders, products, variants, inventory and collections.
- Joins order revenue to Google Ads, Meta and TikTok spend in the same task.
- Ranks products by contribution after acquisition cost, not by gross revenue.
- Reconciles platform-reported conversions against orders that actually settled, including refunds and discounts.
- Drafts store writes — product status, inventory-driven changes — each pausing for approval.
- Creates and retires discount codes, adjusts inventory, edits product copy / tags / SEO and prices, adds product media and manages the app's web pixel — each checking the store's granted scope first and pausing for approval; prices, absolute stock sets, deletions, deactivating a code (by status or by date), a discount of 50 % off or more and status flips wait for a person — and builds UTM-tagged landing URLs for your ads.
- Raises threshold alerts on orders, revenue and AOV per closed shop-local day.
Real use cases
- "Which products lose money once ad spend is counted?" — the single question a store dashboard cannot answer.
- Pause ads for out-of-stock products before the spend is wasted.
- Reconcile a platform's conversion count against settled orders, which is where reported and real ROAS diverge.
- Daily sales anomaly alerts, where a stopped store correctly reads as a zero rather than as missing data.
Multi-store, handled explicitly
Connecting a second store adds it alongside the first rather than replacing it, and reconnecting the same store refreshes it in place so scheduled tasks and alert rules pointing at it stay alive.
With more than one store connected, a write that does not name a store is refused and both are named, rather than defaulting to whichever happens to be first.
Shopify ships its own MCP servers and they are good at store operations. What they cannot see is ad spend — which is the whole reason to connect the store to an ad-ops agent rather than the other way round.
Frequently asked questions
- What can an AI agent do with my Shopify store?
- Read orders, products, variants, inventory and collections, join that to your ad spend in the same task, and draft store writes that pause for your approval. The value is the join — ad spend is not in the Shopify admin.
- Can it tell me which products are actually profitable?
- Yes, once ad accounts are connected — it ranks products by contribution after acquisition cost rather than by gross revenue, which usually reorders them substantially.
- Can I connect more than one Shopify store?
- Yes. Each store is added alongside the others, reconnecting refreshes in place, and a write that does not name a store is refused with both stores named rather than guessing.
- Can it change my store?
- Yes, approval-gated: discount codes, inventory, product copy, tags, SEO and status, variant prices, product media and the app's web pixel, each checked against the scopes your store's authorization actually grants. Deleting a code or the pixel, setting stock absolutely, repricing, deactivating a code (by status or by date), a discount of 50 % off or more and flipping a product's status always wait for a person. No dedicated tool writes orders or checkouts. The Shopify app does not hold the permission to change orders, so through it no order mutation is reachable; on a connection that carries order write permissions, the generic mutation tool reaches any mutation those permissions allow, order and payment mutations included. Refunds always wait for a person — a refund, or a return processed with one — and so do captures, cancellations, voids, completing a draft order, charging a saved payment method, editing an order's line items and every delete mutation; creating an order, updating its notes, tags or addresses, or marking one as paid (which records a payment taken outside Shopify) takes the ordinary approval.
- Do I need an Admin API token?
- No — install the Shopify app from the Shopify App Store (Integrations → Shopify → Install from Shopify): Shopify picks the store, lists the permissions the app asks for and refreshes its access by itself. There is nothing to create, copy or paste.
- What store data does Agent Planners keep?
- The store's domain, name, currency and time zone, the granted permissions and the encrypted tokens; and, inside your workspace, the results of the tasks you run, deleted at your task-history retention period. Shopify does not give the app customers' names, emails, phone numbers or addresses. When you uninstall the app, Shopify's shop/redact request deletes every workspace's connection to the store; see the Shopify guide and the privacy policy for the details.
- How does this differ from Shopify's own MCP or Sidekick?
- They are better at store operations — theme edits, segments, Flow workflows — and we do not compete there. Neither can see a cent of your ad spend, which is the question this integration exists to answer.