You can ask an AI assistant about supported Unleashed records through an MCP connector without first building a dashboard. Begin with one company, one warehouse and one operational question. Verify the returned records before using the answer to make a purchasing or stock decision.
Atso connects business software to AI assistants through MCP, the Model Context Protocol. Its Unleashed connector reads products, warehouses, stock and supported order records. This is useful for an investigation that needs several follow-up questions, while a maintained dashboard may suit a fixed recurring report better.
Which stock questions are good candidates for an AI review?
Choose questions that can be answered from a small, clearly defined set of records. Ask what is present and what needs checking before asking the assistant to decide what the business should do.
Useful starting questions include which products are recorded in a selected warehouse, which sales orders are backordered, and which purchase orders need a buyer’s review. Keep the output tied to product codes and order identifiers so another person can verify it.
A broader request such as “optimise all purchasing” hides many assumptions. The assistant would need to understand lead times, demand patterns, supplier constraints and your service priorities. If those facts are not available, the answer should identify them as missing. It should not substitute a generic stock formula and describe the result as a company policy.
What information does Unleashed expose for stock on hand?
Unleashed’s stock-on-hand interface provides product quantities and related stock information, with filters for the scope of the read. The important point is that a field’s meaning should survive the journey from the inventory system into the chat.
The official Stock On Hand documentation is the reference for the source records. Atso’s selected columns include product code, warehouse code, quantity on hand, available quantity, average cost and total cost. A quantity on hand is not automatically the quantity you can promise to a new customer.
Here is a proposed review checklist:
| Check | Question for the reviewer |
|---|---|
| Company | Is this the intended inventory business? |
| Warehouse | Is the location explicitly selected? |
| Product | Does the code identify the item we mean? |
| Quantity | Is this on hand, available or another measure? |
| Time | Is the request current or for a supported as-at date? |
| Coverage | Were all relevant pages returned? |
Retain the source labels in the working table. Renaming several different quantity fields to “stock” makes the report shorter but can remove the distinction needed to make a sound decision.
Why should you read one warehouse at a time?
An explicit warehouse filter makes the location of the result clear. Atso’s own connector notes that an unfiltered stock read can return an empty warehouse code, so the safer review asks for each warehouse deliberately.
The connector preserves results for each company and does not silently add stock across companies or warehouses. An assistant should preserve those boundaries in its explanation.
List the warehouses first, then use the selected warehouse code for the stock request. The Unleashed Warehouses documentation describes the source records behind that selection. A branch name somebody uses in conversation may differ from the warehouse code stored in the system.
If you need a wider view, repeat the same question for each named warehouse and keep the results alongside one another. Before requesting any aggregate, confirm that the selected datasets are non-overlapping and that the measures mean the same thing. Avoid adding an all-warehouse result to individual warehouse results.
How do sales and purchase orders help explain the stock position?
Order records help you investigate commitments and expected activity around a product. Their status matters as much as the quantity shown.
Unleashed documents its Sales Orders interface, including order records and their details. In Atso, the sales-order reader accepts filters such as warehouse, date range and order status. Keep parked, backordered, completed and deleted orders distinct rather than treating every line as current demand.
For a purchasing question, read the relevant purchase-order records and inspect the details that are actually returned. A placed order is not proof that stock arrived. If the delivery information you need is absent, ask the buyer to confirm it instead of inferring a receipt from the order date.
Consider an invented example: a product has limited available quantity and an open purchase order. That gives you two observations, not a guarantee that customer demand is covered. You still need to establish when the supplier will deliver and which existing commitments the incoming units must satisfy.
What prompt should you start with?
Start with a read-only review that asks for the facts, the source identifiers and the unanswered questions. Keep the first run small enough to check manually.
List the Unleashed companies and warehouses I can access. For the company and warehouse I select, inspect the product codes I provide. Show on-hand and available quantity separately. Find relevant backordered sales orders and open purchase orders using supported filters. Preserve order status and company identity. Flag incomplete reads and missing delivery information. Return questions for our buyer; do not create orders or adjust stock.
This is a suggested prompt, not a recorded customer result. Select a few familiar product codes and compare the answer with Unleashed. Once you trust the field meanings and coverage, widen the review to a larger product group.
MCP provides the connection between the AI application and the server’s tools. It does not make the assistant an inventory authority. The useful part is that follow-up questions can refer to retrieved business records instead of relying on a pasted summary from memory.
When is a dashboard or export the better choice?
Use a dashboard or a maintained export when the job needs a fixed report, a stable historical dataset or repeated analysis at a scale that exceeds an interactive read. Use a conversational review when you need to inspect a bounded question and follow the evidence.
Atso pages through supported resources and applies request limits. Large reads can take several calls or require a narrower scope. A partial result is still useful if it is labelled, but it should not be distributed as a complete stock audit.
You can open Atso to connect Unleashed and test one question. For an introduction to the connection model, read MCP for business software. Finish the first review with a checked answer and a named next action, then decide whether the same workflow deserves a regular report.