TLDR
- Prompt-based product discovery lets shoppers describe what they need in natural language instead of navigating filters.
- MCP gives the assistant a defined set of tools against the real catalog.
- Clean catalog data and clear tool boundaries decide whether it works in production.
What the demo shows
Complex catalogs are hard to browse. Shoppers have to know which filters, categories and option names map to what they need. In this live demo, prompt-based product discovery lets a customer describe what they need in natural language and surfaces the right products, configurations and options, powered by MCP.
Why MCP fits product discovery
MCP gives the assistant a defined set of tools against the real catalog, rather than leaving it to guess from a static page. The server controls what is exposed, so the assistant answers from the same product data your storefront uses. The Model Context Protocol specification describes how tools are defined and called.
What to check before you build this
Is your catalog data clean enough to answer a question like a person would ask it?
Which tools should the assistant have, and which should it never have?
How will you review what shoppers asked, and what the assistant answered?
We learned the production side of these questions by running MCP on our own systems, described in We Run Our Agency on MCP, and in Giving AI Agents Real Employee Identity with MCP OAuth. For the commerce side, see our conversational search and MCP server solutions.