Grietens

How to connect the Excel files, ERP and machines of a factory to Claude or ChatGPT

Updated

The question this guide answers
How do I connect the Excel files, ERP and machines of my factory to Claude or ChatGPT so I can ask questions about production?

Most small factories keep their production knowledge in several places. Orders sit in an Excel workbook or an ERP, cutting lists and stock counts are separate spreadsheets, delivery notes are PDFs, and the machines report nothing or report to software nobody else reads. Claude or ChatGPT can explain a spreadsheet pasted into a chat, but it forgets the sheet in the next chat and cannot join an order to the job that made it.

This guide explains how to give an AI assistant lasting access to these sources through Grietens, a service that collects the files, systems and machine data of a factory in one place so that the owner can ask his own AI about them. It covers the steps, what you can ask, the safety rules, the cost and the limits.

How the connection works

Claude and ChatGPT can use outside tools through the Model Context Protocol (MCP), an open standard for connecting AI applications to data sources. A service publishes an MCP server at a web address, and the AI app adds that address as a connector. The app then asks the person to approve the connection on a page of the service, using the OAuth sign-in described in the MCP specification. From then on, every new chat in that app can call the tools of the service.

The Grietens MCP server lives at https://app.grietens.cloud/api/v1/mcp. Its tools read the data of the factory, add new records, accept files and set up connections to other systems. The data stays in Grietens, and the assistant fetches only what a question needs.

Step 1. Add the connector

In Claude, open Customize (Settings in older versions), then Connectors, then Add custom connector. Name it "Grietens", paste the address above, click Add and then Connect. In ChatGPT, open Settings, then Apps & Connectors, then Advanced settings, turn on Developer mode, click Create, name it "Grietens", paste the address, choose OAuth and click Create and Connect. Menu names change from time to time, so look for the words "custom connector" or "MCP".

A Grietens page opens with two buttons. The button I'm new to Grietens creates a new, provisional workspace with one click and connects the app to it, and no account is needed yet. The button I'm a Grietens customer signs in to an existing account. A connector added during a chat appears in the next chat, not the current one, so open a new chat afterwards.

The provisional workspace is fully usable. When you want your own login, ask the assistant for the claim link. On that page you choose an email and password and accept the data terms. A workspace that nobody claims within 30 days is deleted with everything in it.

Step 2. Name the factory and send a first file

The assistant walks you through five short steps. You name the factory, add the connector, send a first file or list, ask one question, and claim the account when you are ready.

A tool call carries text and not the bytes of a file attached to a chat, so files reach Grietens in one of three ways.

  1. An upload link. The assistant asks Grietens for a signed link that belongs to your workspace, works for 30 minutes and takes up to 25 files by default. In claude.ai or ChatGPT you open the page of the link on a phone or laptop and drop the files in. An assistant that can run commands, such as Claude Code or Claude Desktop, sends each file itself with one curl line.
  2. A public link. If a file is already shared at an https address, Grietens downloads it itself. Private and local addresses are refused.
  3. Text the assistant already holds, such as a table pasted into the chat, which is kept as a document.

Files can be up to 50 MB each. A spreadsheet becomes tables, the text of a PDF is read, and a photo is kept as it is. Each file also becomes a document that can be linked to the order, delivery, customer or machine it is about.

Step 3. Let Grietens sort the data

You do not have to rename columns or follow a template. Grietens keeps every row exactly as it arrived and makes each sheet queryable at once. In the background it maps the columns onto a small shared vocabulary of factory terms, namely orders, jobs, products, customers, machines, quantities, dates and stock movements (received, issued and stock on hand). Columns that fit none of these are kept under their own names, so nothing is thrown away. Problems in the data, such as a date that cannot be read or a row that looks like a sheet total, are recorded as data issues rather than silently fixed.

When the same customer is written two ways, such as "SIA Koks" and "Koks", Grietens notices a likely duplicate. The assistant joins the names or keeps them apart when you say which is right.

Step 4. Connect an ERP or webshop

If your orders live in a system with an HTTPS API, the assistant can set up a connector that pulls new records on a schedule. When the system needs an API key, the assistant gives you a link to a secure Grietens form, and you type the key there. The assistant never sees it. Before a connector goes live it is checked and given a test run that writes nothing. A connector may only read its source and add records to Grietens. It can make GET requests to the hosts it declared and is never allowed to change or delete anything in your ERP.

An app of your own, such as a barcode scanner or a script, can instead send records to a signed push address. Sending the same record twice stores it once.

If your system has no API, export a file from it and send that file as described in Step 2.

Step 5. Connect machines

Machine data needs a Grietens box on site. Grietens Link reads the data a machine itself holds (from a PLC, a CNC controller or a networked device, over Ethernet or RS-485) through a box and never writes to the machine. A Grietens camera box can count machine cycles where a machine reports nothing, and only crops of the machine area are sent for AI analysis. The boxes are new hardware, so check with Grietens whether the controller of your machine is supported before you plan around one. Compatibility with every machine is not promised.

What you can ask afterwards

Once the data is in, questions go to your assistant in its normal chat. Typical questions look like these.

Answers are only as good as the data. A cycle counted by a camera is not a sold or accepted part. The slowest station is a candidate bottleneck, not a proven one, until jobs, queues and delays have been checked. A machine that sends no data is offline, not necessarily idle. Ask the assistant which records and time window an answer rests on, and check important numbers against the source.

Safety

What it costs

Grietens describes its pricing as metered infrastructure. Questions asked through your own Claude or ChatGPT are free and unlimited, because your assistant does the thinking on your own subscription and Grietens only serves the data. Work that Grietens runs in the background on your behalf is metered at what it costs Grietens plus 50%, together with storage and records ingested. That work includes mapping new spreadsheets, maintaining its memory of your orders and machines, repairing connectors, and the chat assistant built into the Grietens dashboard. Connected machines are priced per machine per month. As of October 2026, usage is recorded per workspace and shown on the usage screen, but it is not yet charged, and the free allowance has not been set. The owner can set a monthly spending cap. When the cap is reached, background AI jobs wait, while incoming data is never held back.

Limits

Sources

All guides · Set up Grietens with your AI · Create a factory workspace
Other languages · Latviešu