Shopify AI Agent Tools: What Your Agents Can Actually Do in Your Store

Most "AI for Shopify" is a chatbot bolted to the storefront. Clearly ships something different: a set of store tools your AI agents can call directly — browse the catalog, read orders, pull a health-and-revenue rollup, edit products, and attach generated images — reachable over MCP, the beehaven CLI, and (soon) typed clearly_shopify_* function tools. That's the difference between an agent that talks about your store and one that operates it.

Here's exactly what those tools are, how to call them, and the guardrails that keep a live store safe.

How the connection works: a store is an attachment on a team

Clearly connects a Shopify store to a team workspace — the store rides along as an attachment on the team, not on your personal account. That's deliberate: once a store is attached, every AI agent on that team inherits the store tools, usage meters against the shared team AI pool, and access is governed at the team level.

Under the hood each tool is a store RPC. From the CLI you reach them through the merchant proxy (user:store-rpc); over MCP they surface to your agents as callable tools; and typed clearly_shopify_* function tools are on the way, so an agent gets first-class, schema'd Shopify actions right alongside Clearly's design tools.

The tools

Five tools ship today — three read, two write:

Tool	Type	What it does
list-products-for-editor	Read	Browse the catalog — titles, IDs, status, the fields an editor needs.
list-orders	Read	Recent orders, pulled from the synced order mirror (no live-API round-trip per call).
store-overview	Read	One call → catalog health + an order/revenue rollup. The "how's the store doing?" answer.
update-product	Write	Edit title, descriptionHtml, tags, or status via Shopify's productUpdate. Supports dryRun.
attach-product-image	Write	Attach a hosted/generated image to a product via productCreateMedia. Supports dryRun.

The reads let an agent understand the store; the writes let it change the store — always with a preview available first.

Reading the store (safe by definition)

Connect to the store, then call. From the beehaven CLI it's a two-step: point at the store, then invoke a tool.

beehaven connect shopifyStore/<store-id>
beehaven call store-overview '{}'

store-overview is the one to reach for first — a single call returns catalog health (how many products, how many missing images or descriptions) plus an order and revenue rollup, so an agent can open a session already knowing the shape of the business instead of paging through the admin.

Browsing the catalog is just as direct:

beehaven call list-products-for-editor '{"limit": 20}'

and recent orders come straight from the synced mirror:

beehaven call list-orders '{"limit": 10}'

Because these are reads, there's nothing to approve — an agent can call them freely to build context before it proposes any change.

Editing products — with a dry-run preview

update-product is the workhorse write. It maps to Shopify's productUpdate mutation and edits the fields that matter for merchandising: title, description HTML, tags, status. The important part: every write supports dryRun. Pass "dryRun": true and the tool returns the exact change it would make — the before and after — without touching the live store:

beehaven call update-product '{
  "productId": "gid://shopify/Product/8123456789",
  "title": "Hand-Poured Soy Candle — Cedar & Sage",
  "tags": ["candles", "soy", "cedar", "sage"],
  "status": "ACTIVE",
  "dryRun": true
}'

Review the preview, drop dryRun, run it again, and the change lands. This is the pattern an agent should follow on anything that mutates a live catalog: preview, confirm, commit. It's the same discipline a careful human uses — just enforced as a first-class option on the tool itself.

Generate an on-brand image, then attach it

This is where Clearly's design DNA shows. attach-product-image attaches a hosted image to a product via productCreateMedia — and it pairs directly with Clearly's own clearly_generate_image tool. So a single agent can, in one session:

1. Generate an on-brand product image with clearly_generate_image,

2. Take the resulting hosted URL, and

3. Attach it to the product with attach-product-image — previewing with dryRun first.

   beehaven call attach-product-image '{ "productId": "gid://shopify/Product/8123456789", "imageUrl": "https://files.clearly.sh/generated/cedar-sage-candle.png", "altText": "Cedar & sage soy candle on a linen backdrop", "dryRun": true }'

No hand-off to a separate design tool, no manual upload, no download-then-reupload dance. The agent that makes the image is the agent that ships it onto the product.

The guardrails

An agent that can touch a live store needs a leash. Today's guardrails:

• Dry-run on every write. Preview the exact change before it lands — no blind mutations against a live store.
• Ownership-verified store. You can only attach a store you actually own: you must have installed the Clearly Shopify app on it. There's no back-door access to someone else's shop.
• Deny-filtered actions. The merchant proxy allow-lists what agents can call. Admin, credential, raw SQL, export, and delete actions are denied — an agent gets the merchandising tools, not the keys to the vault.
• Metered usage. Every call meters against the team's AI pool, so store activity is accounted for like any other agent work.

The through-line: agents get the tools to run the store, and none of the tools to compromise it.

How to connect

1. Create a team. Store tools live on team workspaces, so start by making a team.
2. Install the Clearly app on Shopify. This is the ownership handshake — it grants Clearly scoped access to your store.
3. Attach your store by domain. Point the team at your myshopify.com domain, and the store becomes an attachment on the team.

From that moment, every agent on the team can call the tools above — from the CLI, over MCP, or (soon) as typed clearly_shopify_* tools.

Want to watch the agents work? Mwah puts a face on each live agent on your desktop, so "an agent is editing my catalog right now" is something you can see, not just trust.

FAQ

Do agents get raw admin access to my store? No. Access runs through a deny-filtered merchant proxy: merchandising reads and writes are allowed; admin, credentials, SQL, export, and delete are blocked. Agents get a curated toolset, not the Shopify admin.

Can I preview a change before it goes live? Yes — every write tool (update-product, attach-product-image) supports dryRun. Pass "dryRun": true to see the exact change without applying it, then re-run without it to commit.

Where do the orders come from — the live API on every call? From a synced mirror of your orders, so list-orders and store-overview are fast and don't hammer the Shopify Admin API on every read.

Is the store tied to my personal account? No — a store attaches to a team workspace, and usage meters against the team AI pool. That's what lets a whole team's agents share one connected store.

What can I call these tools from? The beehaven CLI (user:store-rpc) today, MCP for your agents, and typed clearly_shopify_* function tools soon.

Point your agents at your store

The tools are real and shipping: browse the catalog, read orders, roll up revenue, edit products, and attach generated images — all agent-callable, all guard-railed, all with a preview before anything writes. Connect a store to a team and your agents stop describing your shop and start running it.

See it → Clearly's Shopify integration.

---

Related: Manage your Shopify store with AI agents · Claude Code for Shopify