Košík.cz is live on the ChatGPT App Marketplace and, according to BigHub's case study, is the first e-grocery retailer in Central & Eastern Europe to offer a fully transactional AI agent. Customers describe what they need in plain language and the AI handles the rest. Product Director Jan Kopecký predicts that within the next few years approximately 10% of all Košík.cz orders will be placed through AI agents, and that by 2030 this share could grow to as much as 25%.
"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, Product Director at Košík.cz.
Online grocery shopping solved logistics, but customers still browse categories, scroll product lists and manually build their carts, which is slow, repetitive and uninspiring. Rather than building another recommendation chatbot that suggests recipes and sends customers back to the website, Košík.cz wanted an AI agent that actually completes the purchase. The project took approximately 3–6 months from kickoff to launch.
The Košík Shopping MCP connects ChatGPT directly to the Košík.cz e-commerce infrastructure and executes five steps. The agent interprets natural language into structured shopping context, queries the full Košík.cz catalog, which BigHub describes as 4x the range of a typical supermarket, assembles a cart for the customer's review and approval, sets the address, delivery window and payment method, and then confirms the transaction and triggers fulfillment without the customer leaving the conversation. The hardest part was the semantic layer: a deep semantic mapping of the product catalog so the model understands relationships between products, categories and human intent, plus a data pipeline that keeps inventory and pricing real-time. "The AI is actually better at searching our catalog than traditional full-text search. Because it understands context — it understands what you mean, not just what you type," said Kopecký.
The solution is built on the Model Context Protocol (MCP), chosen over a closed proprietary integration to keep the architecture open to other AI platforms. The technology stack is MCP, the ChatGPT Apps SDK, Docker and Kubernetes. BigHub designed the architecture, implemented the MCP server, built an authentication layer mirroring Košík.cz's existing auth, developed UI widget components with the ChatGPT Apps SDK and delivered Docker containers. Košík.cz provided API access and documentation, defined the shopping flow and business logic, handled the transaction and payment layer, deployed on its own Kubernetes cluster and owns the customer relationship and support.
Looking ahead, Kopecký calls the next phase "agent-to-agent commerce," in which a customer's personal AI assistant communicates directly with Košík's commerce agent.