Košík.cz, the Czech online grocery leader, became the first grocery retailer in Central & Eastern Europe to go live on the ChatGPT App Marketplace, deploying an MCP-based AI shopping agent — built by BigHub — that lets customers complete a full grocery order through a single conversation, with Košík.cz CEO Jan Kopecký predicting that 10% of all Košík.cz orders could come through AI agents within a few years and 25% by 2030.
"We weren't interested in building another chatbot that just talks about food. The breakthrough is giving the agent hands — letting it reach into our warehouse, see real prices in real time, and actually reserve products for the customer," said Jan Kopecký, CEO of Košík.cz.
The deployment addressed a shopping-experience gap rather than a logistics one. Online grocery had solved delivery, but the act of shopping itself — browsing categories, scrolling product lists, manually building carts — remained slow, repetitive, and uninspiring, especially for the routine weekly restocking most households do. Košík.cz wanted not another recommendation chatbot but a transactional AI agent that actually completes the purchase.
Košík.cz brought the commerce vision, the warehouse and checkout infrastructure, and the product catalog. BigHub contributed the AI architecture, the Model Context Protocol (MCP) implementation, and the integration layer connecting it all to ChatGPT. The project ran roughly three to six months from kickoff to live deployment.
Technically, the "Košík Shopping MCP" executes a five-step shopping flow on top of ChatGPT: the agent interprets natural-language intent into structured shopping context, searches and matches against the full Košík.cz catalog — roughly four times the range of a typical supermarket — assembles a cart for customer approval, configures delivery and payment, and executes the order, all without leaving the conversation. The hardest part, per BigHub, was building the semantic layer mapping products, categories, and human intent, plus the real-time inventory and pricing pipeline needed to avoid promising out-of-stock items. MCP was chosen as an open standard so the architecture can extend to other AI platforms.
At operational scale, the agent is positioned as a new sales channel rather than a feature. Kopecký's prediction — 10% of Košík.cz orders via AI agents within years, up to 25% by 2030 — sets a financial target. The pitch rests on a behavioral assumption: grocery shopping is a utility, not a pleasure, and once customers experience conversational restocking, they are unlikely to return to manual browsing.
Looking ahead, Košík.cz and BigHub describe the next phase as "agent-to-agent commerce," in which a customer's personal AI assistant communicates directly with Košík's commerce agent — checking the fridge, knowing about allergies and dinner guests, negotiating prices, booking a delivery slot, and confirming with a single notification.