Shopify

Two nodes, one connected store. The action writes into the store — add or update a customer, tag an order, move stock — or reads it back to enrich a payload. The trigger fires a flow when something happens in the store: an order comes in, a product changes, stock moves. The action can also be handed to an AI Node as an AI toolkit.

Connect a store

Both nodes use the same credential, and you connect it once. Click Connect Shopify in either node's credential picker, or from Settings → Credentials.

The store travels with the credential from then on. That is why the trigger has no “which store” step and the action node shows no store field: one credential is one store. To work with a second store, connect it as a second credential.

The action node

Twelve operations, all on Shopify's GraphQL Admin API. The API has hundreds of mutations; these are the ones people actually automate — customers, tags, stock and lookups, the order lifecycle (draft, complete, fulfil, cancel), metafields and new products.

Add or update customer

Looks the customer up by email, then creates them or updates the fields you set. Point the email at your incoming payload — {{payload.email}} — and leave blank anything you do not want to touch: empty fields are left exactly as they were on an existing customer.

Add or remove tags

Tags an order, a customer, a product or a draft order. You pass the Shopify global id — gid://shopify/Order/1234 — and a Shopify trigger hands you one directly in {{payload.body.admin_graphql_api_id}}. The id has to be of the kind you picked under What to tag: a customer id with Order picked is refused instead of tagging the customer. Tags can be typed vip, mx or arrive as a list from the payload.

Adjust inventory

For the case where the real stock lives in another system. You give it an inventory item, a location, and how much to move.

The inventory item is not the product and not the variant — a variant points at one. And the reason you type lands in the store's own inventory history, so someone reading it in Shopify later can tell why the stock moved.

Find orders, customers or products

Reads the store to enrich a payload that came from somewhere else. Takes Shopify search syntax — email:ada@example.com, financial_status:paid — and returns one record by default, up to 50 orders or customers, or 25 products. Products come with their variants — the variantId a draft order's line items need and the inventoryItemId inventory needs. Newest first sorts by creation date, and when there is more a nextCursor comes back to pass as Next page.

Get order and List locations

Get order reads one order whole: line items, payment and fulfilment status, customer, shipping address and totals. List locations returns the places the store keeps stock, with the ids Adjust inventory needs.

Draft, complete, fulfil, cancel

Create draft order builds an order the merchant can review or invoice, and charges nobody. Its line items are a JSON list — type it, or point it at {{payload.items}} when the payload already has one. Complete draft order turns it into a real order, Fulfil order marks it shipped and Cancel order cancels it; none of the three is undone by running it again. Set a metafield writes custom data onto an order, customer, product or variant, and Create product puts a product in the catalogue.

AI toolkit mode

Once the store is connected, the node asks for a Mode. AI toolkit turns it into fifteen tools that an AI Node — or an MCP client, through one of your MCP servers — calls by name.

A search without query lists without a filter. Products come with their variants, so the model can go from a search to a draft order or an inventory move without asking you for ids.

What you control

• Blocked from the start. cancel_shopify_order, complete_shopify_draft_order and fulfill_shopify_order move money, stock or a shipment and can't be undone, so they start blocked. Unblock them under Block these tool calls if the agent should.
• Emails to customers. Whether fulfilling or cancelling emails the customer is yours: off by default, and no tool has an argument for it.
• Restocking. Whether a cancellation puts the stock back is yours too, off by default.
• New products. Born as Draft, hidden from shoppers, unless you pick Active.
• Marketing consent. The model never changes a customer's email marketing consent, and it never picks a payment gateway — which means a draft it completes is marked as paid. That is why complete_shopify_draft_order starts blocked.
• Characters per call sent to the model. 20,000 by default, anything from 1,000 to 200,000.

What goes back to the model

The same response as the operation. Lists that don't fit the characters limit are cut whole items at a time, with a notes field saying how many were left out. Arguments are used as written: a {{payload.x}} the model writes is not resolved against the flow.

Testing a tool

{ "operation": "search_shopify_products", "query": "title:snowboard*", "limit": 5 }

The trigger

Pick a store credential, then pick what fires it. Six buckets: orders, fulfilled orders, products, customers, inventory levels and checkouts. Each expands to Shopify's own webhook topics when the trigger is activated.

There are no per-event filters here on purpose. Everything of the kinds you picked arrives, and a Filter node right after the trigger narrows it down — only paid orders over a certain amount, say.

If the app is removed from the store

Every Shopify trigger also listens for the app being uninstalled, whether you asked for it or not. It is the only signal Shopify gives that the credential has stopped working — without it, an uninstalled store looks exactly like a store where nothing is happening.

That event never runs your flow. It marks the credential as needing a reconnect, so the screen can say so instead of leaving your nodes failing with an error that looks like a network problem.

What the trigger sends downstream

The delivery metadata is flattened, and Shopify's raw body is kept whole under body so nothing is lost — every topic has a different shape and flattening forty of them would be inventing a contract:

{
  "id": "b4f1c2e0-...",
  "topic": "orders/create",
  "shop": "your-store.myshopify.com",
  "eventId": "9f3a...",
  "triggeredAt": "2026-08-25T18:04:11Z",
  "apiVersion": "2026-07",
  "resourceId": "gid://shopify/Order/1234567890",
  "body": { "...": "the raw Shopify body" }
}

Reach anything Shopify sent with {{payload.body.whatever}} — the line items of an order, the variant of a product, the tags already on a customer.

Trying it out

The Test push button reads the most recent orders, products, customers and stock levels of your store and shows them the way a delivery will arrive, without running the flow. It is rebuilt from the store's current data, so a live delivery can carry more fields and its dates come in the store's time zone. To see a real one, activate the trigger and do the thing in your store — placing a test order is the quickest.

Good to know

The rate limit counts cost, not calls

Shopify meters the GraphQL Admin API by how expensive each query is, not by how many you make: 100 points per second on a standard plan, 200 on Advanced, 1000 on Plus. A search asking for fifty records costs more than one asking for one. If the store throttles, HostWebhook backs off and retries once.

A trigger can go quiet without saying so

The API version moves quarterly

HostWebhook pins one Admin API version and updates it deliberately. Shopify supports each version for at least twelve months, so nothing breaks under you between updates.

One app per store, for now