What Is the Linnworks MCP Connector?The Linnworks MCP Connector gives AI tools controlled, read-only access to operational data stored inside Linnworks.Instead of manually combining exports from orders, inventory, channels, warehouses, purchase orders, and customer records, users can ask questions in natural language and retrieve the relevant data from the connected Linnworks account.The flow is straightforward: Linnworks → Read-Only MCP Layer → AI Interface → Business AnswerThe connector exposes approximately 60 read-only tools across roughly 134 business tables, while typed parameters and validation rules control how data is queried.
Step 1Ask Complex Linnworks Reporting Questions in One QueryDate, channel, SKU, and location, in the same question. Linnworks reporting filters one dimension well. Real questions carry four at once, which is why they end up in a spreadsheet. This reads across roughly 134 business tables directly, so the combination is one question rather than four exports and a lookup. Open order queues by location, orders aged into buckets, top SKUs this month, revenue split by channel, what is on order and when it lands. All of it in the same place you asked the last question. Long answers page properly. Ask for the next page, and it continues exactly where the last one stopped, no rows skipped, none repeated.This approach is particularly useful when Linnworks sits at the center of a multi-system ecommerce environment where orders, inventory, customers, and financial data need to remain connected across platforms.
Step 2A Read-Only Linnworks MCP Connector, By DesignNo path in this connector changes anything in Linnworks. Not disabled, not built. No create, cancel, refund, re-price, re-locate, or re-address. Nothing to misconfigure, and no permission to get wrong. So you can point it at the live account on the first call. Most connectors start on a copy of the data because the risk of a write is real. Here, the risk doesn’t exist, and the evaluation is over in an afternoon instead of a fortnight. The model never writes a query against your database. It passes typed parameters; the server builds the query behind a validator that refuses anything it does not recognize. User and permission tables are blocked outright. Whatever is asked is unreachable.Connect the accounts you need and pick a default; “Remove” is the only control that changes anything, and it only unlinks.
Step 3It Knows Where the Numbers LieThe Linnworks MCP connector accounts for reporting behaviours that can otherwise distort multi-location inventory and payment-date reporting.Two details that quietly break Linnworks reporting, handled before you ask. Stock is never summed across locations. Amazon FBA and fulfillment-center locations mirror availability rather than holding their own; add them to a warehouse figure, and you have counted the same stock twice. Answers stay per location and key on the location’s ID, because two locations can share a name.We have handled similar inventory challenges in production Linnworks environments. See how a Linnworks and BigCommerce integration automated stock management across more than 9,000 items. The payment date isn’t on the order header. It lives on the order’s additional information, which is why revenue-by-paid-date reports built against the obvious field come out wrong and nobody notices for a quarter. Neither of these is in the documentation you would read first. You learn them by getting a report wrong once, and they are already handled here. Stock reported per location, and deliberately never totaled.
Ten things to type on day oneThese are the prompts we use in the demo. Nothing is rephrased for the website. “Open order queue”: unprocessed orders for one location “Orders received last week”: any channel, in a date window “Aging unshipped orders”: open orders bucketed by age “Show order #204418 with lines”: one order and its line items, in a single call “Top SKUs this month”: best sellers for the current month “Revenue by channel”: orders and revenue split by sales channel “Stock at the Leeds warehouse”: per-item stock at one location “What is low or out of stock?” : at or below reorder point, and at zero “Incoming purchase stock”: what is on order and when it lands “Find a customer and their orders”: customer lookup with order history.
Under the hood60 read-only tools across roughly 134 business tables. Typed parameters behind a fail-closed validator; the model never writes a query. Keyset pagination, so pages never skip or overlap. Stock reported per location, never summed. User tables blocked.MCP is also becoming increasingly relevant to finance and accounting automation, giving AI agents structured ways to access business-system data without relying on repeated exports and uploads.Point it at the live account. It cannot break anything.