Conversations server
What is the Conversations server?
The Conversations server is a Loomi Connect MCP server for prototyping AI shopping experiences. Connect your AI assistant to it and the assistant can search a sample product catalog, refine results mid-conversation, and guide undecided shoppers, all in natural language.
Use it to explore what conversational commerce on top of Loomi Connect looks like — prototype an AI shopping assistant, build a chatbot demo, or test an agent design before committing to a production integration.
NoteTo prototype against your own catalog, contact the Bloomreach team for a customized MCP server.
What's in the catalog
The server is bootstrapped with the Pacific Apparel catalog — a mid-market, multi-brand fashion catalog with around 13,000 SKUs across women's and men's apparel, footwear, accessories, and jewelry. Around 370 brands are represented, including house brands Roadster and Pacific, plus Adidas, Levi's, MANGO, and others. Most products fall in the $20 to $60 range, with premium items up to $300+.
The catalog is fixed — you can't add products, change prices, or modify attributes. For understanding the product catalog, and more on what to expect when designing prototype queries, see Best practices.
How it works
The Conversations server runs as a remote MCP server over Streamable HTTP, so there's nothing to install locally. Your AI assistant connects to it the same way it connects to any other MCP server, then calls its tools during a shopping conversation.
What your AI assistant can do
NoteSee a working reference implementation at hackathon.bloomreach.works. Log in with the username
hackathon, and passwordhappyhacking. It's a generative UI chat built with Tambo on this MCP server, rendering product grids, modals, and filter panels as React components stream in.
Once connected, your AI assistant can:
- Search the catalog by attributes like product type, color, price, size, category, brand, and more.
- Refine searches mid-conversation as the shopper adjusts their needs.
- Guide undecided shoppers through follow-up questions, then search once it has enough context.
- Suggest ways to narrow results based on the catalog itself, not hardcoded text.
The AI assistant decides which approach to take based on the shopper's input. You don't configure this — it happens automatically.
| Shopper says | What the AI assistant does |
|---|---|
| "Show me women's slim-fit jeans under $40." | Searches the catalog with the right filters and returns ranked results |
| "I need a birthday gift for my husband, he likes cycling." | Asks follow-up questions, then searches once it has enough context |
| "Those are nice, but do you have them in navy?" | Refines the previous search and returns updated results |
| "Anything on sale for Valentine's Day?" | Filters on the seasonal promotion tag and returns matching products |
For more details on testing the server's capabilities, refer to the Test conversations server.
Updated about 2 hours ago