AI assistants (MCP)
Intro
AI assistants lets an assistant such as Cursor, Claude, or ChatGPT read and reason about your store over MCP (Model Context Protocol): explain your automation rules, conditions and filters, run a described or real order through the real engines, work out why a shipment was or was not split or tagged, and price a cart against the checkout shipping rates. It uses the same engines as the Automation Rules Tester, the Rate Tester and Diagnose on an order.
Open the user menu → Settings → AI assistants. Any signed-in member of the store can create tokens; each token is personal, bound to the active store, and limited to that person's own permissions.
Purpose
- Ask "why didn't order #152211 get the
-SDCtag?" and get the rule-by-rule answer - Simulate a rule you are about to build against a real order before you create it
- Let an assistant read a shipping-rate profile back to you in plain English
- Give a developer or consultant scoped, revocable access without a StoresBoost login
What the assistant can do
Access is granted per module; the assistant only sees tools from the modules on the token, and only those your own permissions allow.
| Module | Tools |
|---|---|
| Core | Who am I, store overview, store settings |
| Automation rules | List and explain rules, conditions and tags; dry-run described or real orders (optionally only a draft group); diagnose why an order did or did not trigger rules |
| Import & shipment filters | List and explain store and shipment filters; dry-run them against a real order |
| Orders & shipments | Search and inspect orders and shipments; diagnose why no shipment exists; read and (with changes enabled) edit the merge rules and Merge Settings including the Merge Conflict Tags; list and cancel shipments left with no items; re-sync a fulfilled order's shipments with the store |
| Shipping Rates (checkout) | Read, explain and price-test profiles — zones, groups, tiers, combination and fold rules, triggers |
| Delivery Calendar (checkout dates) | Calendars, rules, exceptions and limit rules; replay a cart to see which dates a buyer gets; diagnose where an order's delivery date came from; capacity per day |
| Operations (alerts, routines, hooks, printing, integration health) | The alerts inbox by level, code or subject — and, with changes enabled, acknowledge, snooze or resolve them; routines and their runs step by step; hooks — list them, read a run, create one as a draft, test it on a shipment, and switch it on (activating a live hook comes back as a preview); document templates and the print history of a shipment; a one-call health check of Shopify webhooks, the checkout rate callback, the fulfillment provider and the weather feed; the log of recent checkout rate requests |
Whether an assistant may change anything depends on how your StoresBoost is configured:
- Read-only — the page shows Read-only for now: assistants inspect, explain, simulate and diagnose, but no tool creates, edits, deletes or activates anything. Rules, filters, shipments and shipping rates are only changed in StoresBoost itself.
- Changes enabled — the modules also carry the tools you have in that area: tag, split, merge, hold and re-run automation on shipments; create and edit rules, conditions, tags and filters; edit shipping-rate profiles. Anything that touches live data comes back to the assistant as a preview you must confirm, and new rules, conditions and filters are created as inactive drafts so you can dry-run them before switching them on.
Every call is logged under your account either way.
Create a token
- Click New token.
- Name it after where you will use it (Cursor on laptop).
- Tick the Modules the assistant may access (all are pre-ticked).
- Optionally set Expires in (days) — 1 to 730; leave empty for no expiry. You can revoke at any time.
- Create token. Copy it now — it is not shown again.
Connect your assistant
The Token created step offers ready-made connections:
| Client | How |
|---|---|
| Cursor | Add to Cursor opens Cursor with the server (storesboost-{store}) and token pre-filled — confirm and you are done. If nothing opens, Copy mcp.json and paste it into ~/.cursor/mcp.json |
| Claude Code | Copy the single claude mcp add --transport http --scope user … command into a terminal (--scope user makes it available in every project) |
| Claude Desktop | Copy the config block into Settings → Developer → Edit Config (claude_desktop_config.json), then fully quit and reopen Claude. It bridges through mcp-remote, so Node.js must be installed |
Other MCP clients: use the Server URL shown on the page with the token as a Bearer Authorization header. Each store gets its own server name, so a multi-store user does not overwrite one connection with another. ChatGPT and Claude's custom connectors need OAuth sign-in, which is not available yet.
Manage tokens
Your tokens for {store} lists each token with its modules, when it was created, last used and expires. Revoke cuts access immediately — any assistant using that token stops working on its next call. Tokens honour exactly the expiry you set; a token without an expiry keeps working until you revoke it.
Troubleshooting
- The assistant says a tool is missing → the module is not on the token, or your own role lacks the matching permission (for example View automation rules). Create a new token or ask an admin to adjust your role.
- "Not authenticated" after it worked before → the token expired or was revoked; create a new one and update the client.
- It sees the wrong store → tokens are bound to the store that was active when you created them. Switch stores and create another token; each store gets its own server entry.
- A change was not applied → in read-only mode nothing is written; with changes enabled, every live change needs your confirmation in the chat, and rules or filters are created as drafts you still have to activate.
Related
- Automation rules — the tester the assistant reuses
- Store Filters & Shipment Filters — filter dry runs
- Shipping Rates · Delivery Calendar
- Roles & permissions — what bounds every token