After the launch of the MCP server at Silpo, Ukrainian IT professionals have already begun testing this scenario in practice. Artificial intelligence has proven to be a good shopping companion.
After the launch of the MCP server at Silpo, Ukrainian IT professionals have already begun testing this scenario in practice. Artificial intelligence has proven to be a good shopping companion.
Senior AI Solution Architect Oleg Yashchuk and R&D Product Manager Leonid Pavlovsky shared their experience of placing orders through AI agents on LinkedIn.
Yashchuk said that his family often uses home delivery of groceries. At the same time, he himself does not like choosing products from the online store’s large catalog — so much so that he jokes that his responsibility was usually limited to receiving packages from the courier and delivering them to the kitchen.
Now he has entrusted this part of the work to AI.
For the first test, Yashchuk asked Claude: «Suggest something interesting to make for Friday dinner. A simple but tasty recipe. Pick the products. And use the appropriate promotions.»
In about 20 minutes, Claude reviewed the recipes, identified the necessary ingredients, checked their availability and current promotions, and added the necessary products to the cart on the Silpo website.
As a result, the AI offered pasta with shrimp in a tomato-cream sauce with step-by-step cooking instructions, and also created a shopping cart ready for payment. All that was left for Yaschuk to do was confirm the order on the website.
«Isn’t this great?» he responded to the test result.
Yashchuk notes that Silpo is probably one of the first in Ukraine to implement a similar scenario through MCP. In his opinion, MCP can become an important technology for the future of retail and customer service, as it allows AI agents not only to make recommendations, but also to interact with external services.
Another IT expert, Leonid Pavlovsky, tested a more complex scenario.
He made his first order through MCP Silpo and called it perhaps the «most LinkedIn-like» supermarket shopping experience. According to him, the system did exactly what he expected: it took into account his list of past purchases, put together a new balanced basket for the week, and also took into account promotions and special offers.
The AI automatically added up to 22 items to the cart. Pavlovsky then added a few more items manually — and the order was ready.
There was also the first test of human-agent interaction. One item was out of stock, but somehow it ended up in the cart anyway. After that, the operator called Pavlovsky and asked what replacement to make.
«Is it correct to say human in the loop?» he joked.
According to Pavlovsky, MCP Silpo also works great in Codex, although the store’s website does not have separate instructions for such a scenario.
He believes that this method will not be used by the largest part of the audience. At the same time, he believes that among those who try AI-purchasing, retention can be high.
Pavlovsky himself plans to continue using the service. For his next order, he wants to set up a separate project with his own preferences, product variations, and regular orders.
Yashchuk prescribed an algorithm for users on how to connect the service.
1. in Claude: Settings -> Connectors -> Add -> Add custom connector -> Name: Silpo -> URL: https://mcp.silpo.ua/mcp -> Connect -> Login to the site
2. in ChatGPT: Plugins -> ⛭ top right -> MCPs -> Add server -> Name: Silpo -> Type: Streamable HTTP -> URL: https://mcp.silpo.ua/mcp -> Save -> Authenticate -> Login to the site
The technological basis of this scenario is MCP (Model Context Protocol) — a protocol that allows AI models to connect to and interact with external services. In the case of Silpo, this means that Claude, Codex and other compatible tools can work not only as consultants, but also perform specific actions in the online store.
