Pickware AI Unveiled: AI Chat and MCP Server in the ERP

p>On June 9, Pickware presented its new AI features in a webinar. According to Pickware CEO Samuel Vogel, it was the largest webinar in the company's history. And the strong interest from the many attendees was no coincidence: with Pickware AI, the Darmstadt-based company introduced a new way of working with the ERP. The name Pickware AI covers two components: an AI chat built directly into the administration and an MCP server that connects the ERP to assistants such as Claude or ChatGPT. As a Pickware partner, we naturally took a very close look at what the new features can do and who they are available to.

What is it about?

Pickware is the ERP for Shopware and Shopify and is used by well over 1,000 merchants. Until now, AI tools and inventory management have mostly existed side by side: product descriptions and images have been generated with AI for some time. However, the AI had no access to the actual data in the ERP, such as orders and stock levels. Pickware now aims to close this gap. Pickware's promise: instead of clicking through menus and data tables, you as the user simply tell the system what should happen.

Incidentally, the idea is not entirely new at Pickware. According to the team around CEO Samuel Vogel, they had already built an initial prototype of the chat in summer 2025, but then put it on hold. It was only the model generation released at the end of the year that tipped the balance in favor of implementing the topic in full.

Pickware AI consists of two components: the AI chat in the administration and the MCP server. But let's take them one at a time.

The AI chat: asking questions where the data lives

The chat can be accessed from any page via the header bar of the administration and works contextually: if an order or product is currently open, questions automatically refer to it. In the live demo, for example, the chat checked the shipping status of an order down to item level, including stock levels, and found the associated open reorder with its delivery date. The chat then also drafted the reply email to the customer.

A second example: a supplier announces a 10 percent price increase. When asked, the chat calculated which of that supplier's products would consequently fall below a minimum margin of 25 percent. Until now, analyses like this were manual work involving exports and spreadsheets.

As an administrator, you always retain full control: you manage who may use the chat via user roles, separately for the chat and the MCP server. And if you're ever not satisfied with an answer, you can rate it directly in the chat. According to Pickware, this feedback feeds into the further development of the AI features. Specific information on the models used was given in the live demo: in the background, it is mainly ChatGPT running via a third-party provider, with servers located in Europe and without the data being used to train the models.

The MCP server: the connection to Claude, ChatGPT and Gemini

MCP stands for Model Context Protocol, a standardized interface through which AI applications access external systems. Pickware has now built in this interface. This allows the ERP to be connected to Claude, ChatGPT or Gemini and combined with other services already connected there, such as an email inbox or Slack. Setup is done via the Pickware settings: there, a server URL is copied and entered in the respective AI assistant; instructions are available in the Pickware help center.

The second demo showed what this looks like in practice. Claude searched the connected Gmail inbox for an expected incoming invoice and matched it to the corresponding supplier order in Pickware. When comparing the line items, it turned out that the quantities and items matched, but the unit prices differed by ten cents per piece. Claude then created an Excel list of the discrepancies on instruction. With two line items, you would probably spot something like this yourself; with 200 line items, it becomes tedious without support. Since the assistants also run as smartphone apps, the whole thing works on the go, with voice input if desired.

A merchant reports

Pickware tested its new AI features with selected merchants during a beta phase. In the webinar, Daniel Kiefer, managing director of the grill brand Barbossa, had his say. As an Enterprise customer, he has already been using Pickware AI for several weeks, including for monitoring partial deliveries:

“Just because I order something doesn't mean I get it. With Pickware AI, I can get an overview of which goods are still inbound and compare that directly with the manufacturers' cancellation lists.”

Daniel Kiefer, Barbossa, in the Pickware webinar

Availability: who gets what?

Pickware AI has been enabled for all Pickware customers since June 9, 2026. However, the range of features depends on the plan booked:

Pickware does not list prices on the product page. Features of higher tiers can be tested free of charge for 30 days.

Very important for Shopware merchants: while Shopify users received the features automatically, Shopware requires an update of the Pickware plugin. In addition, the shop must be running on the current Pickware licensing model. Anyone who still has an older contract should clarify this point before updating.

Outlook: agents and custom routines

In the webinar, Pickware also talked about the next steps. Autonomous agents are already being planned. The agents are intended to make suggestions on their own initiative, for example when a supplier order is 30 days overdue. Once approved, they are to handle routine tasks independently in the future. Pickware also intends to bring the capabilities of Claude Code into the ERP, so that merchants can set up permanent, custom routines via chat instructions. As an example, the webinar cited a real-world request: stock with an approaching best-before date should automatically be separated out 60 days before expiry and listed at a discount in the shop. Also in the works are receiving emails directly in Pickware and stock transfer orders between warehouses.

Our assessment

Pickware itself tempers expectations on one point. Pickware AI is “not magic,” according to CEO Samuel Vogel, but works on the basis of the existing functions and data. If a basic function is missing in the ERP, the chat won't change that either. This assessment is accurate, and it sets the presentation apart from some of the AI marketing of recent months.

To get started, there is a good case for beginning with read-only use cases such as analyses and status queries. Write permissions via the Advanced plan can always be added later. Regarding data protection, the statement that requests run via a third-party provider with servers in Europe and are not used for training is a useful basis. It does not replace a data protection assessment for your own shop; if in doubt, the topic belongs with your data protection officer.

We have been using Pickware in client projects since 2015 and, on request, take care of the plugin update and setup, including a review of the licensing model and user roles. If you have questions about Pickware AI in your Shopware shop, get in touch with us.

Sources:

• Pickware: Webinar “Pickware AI” from June 9, 2026 (YouTube): youtube.com/watch?v=IMlOBBoDsn0
• Pickware: Pickware AI product page: pickware.com/de/pickware-ai