Your AI Now Knows Your Warehouse, Too

In plain language, with the assistant you already use

Warehouse operators connect ChatGPT, Claude or their own, locally operated LLM to their Warehouse Execution System and ask their questions the way they would ask a colleague. There is no new program and no new interface to learn. The interface is the assistant they are already working with.

An example

Asked “Why didn’t the material for order 4711 arrive at the line yesterday?”, the assistant works out for itself which data it needs. It queries, for example, order data, material requirement, stock, movements in the relevant time window and vehicle availability in turn. Five read-only queries that nobody defined in advance.

The answer: the three pallets required were in stock, but could not be moved between 14:32 and 15:18. Two of the three vehicles responsible were unavailable.

An answer for which no dedicated report is worth building, because it comes out differently every time.

From the shopfloor to the CEO

The questions differ depending on who is asking. At the forklift it is about one specific object: where is the pallet, what was left over from the last shift. In warehouse management it is about causes and patterns: which storage locations haven’t been touched for months, which zones cause the most empty runs. At management level it is about development and evidence: how has throughput per vehicle changed over a year, where was a particular batch from goods receipt to loading.

All three access the same data and get the answer in the form they need it. Because the assistant supplies not only the figure but the presentation as well: a table, a chart, a summary for the Monday meeting, a whole dashboard on request. Analyses emerge in conversation instead of as a request to your own IT department.

The division of labor stays clear: the Warehouse Execution System supplies the data, the connected assistant handles the presentation.

A deliberate decision: read-only

identpro has limited access to read-only queries. No assistant can change data, create orders or trigger movement commands. Write functions are not technically provided for. This is not a gap but a division of labor: execution is what the Warehouse Execution System is responsible for. That is what it is built for. The assistant asks; control stays with the system.

Dr. Martin Welp, Managing Director of identpro Systems GmbH:

“The industry is currently promising autonomous agents in the warehouse. We believe the first useful step is a different one: letting people ask their warehouse questions. We deliberately do not let the software act. Anyone responsible for a running warehouse operation wants to see first that the answers are right.”

What warehouse operators need to know

The data does not leave the system: no copy is created, it is read at runtime. Which assistant is connected is for the operator to decide: a locally operated LLM processes locally, a cloud service in the cloud. Access runs through an API key that the operator issues and can revoke at any time. On-premises operation is possible.

There is no customizing lead time. The prerequisite is an existing installation of the Warehouse Execution System.

The technology behind it: Model Context Protocol

Access is made possible by the Model Context Protocol (MCP), an open, vendor-neutral standard for how AI systems access software and the data it holds. It has been maintained since 9 December 2025 by the Agentic AI Foundation under the umbrella of the Linux Foundation, supported by Anthropic, OpenAI, Google, Microsoft and Amazon Web Services, among others.

For warehouse operators this means independence above all: the connection is not tied to one AI provider. Switching assistant does not mean rebuilding the connection.

Warehouse systems making their data accessible to AI assistants is likely to become the norm in the medium term. For warehouse operators the question is less whether they need it – than when.