The Dealer's Desk

Inspiration

Buying a car rarely begins with a perfect form. It begins with a thought: “I need an automatic SUV, I have a trade-in, and I do not know where to start.” The Dealer's Desk turns that thought into structured, inspectable progress without hiding the human decision at the end.

What it does

The experience combines a natural-language agent brief with a clear showroom UI. An agent can call four page-level WebMCP tools:

• search_inventory finds matches in a synthetic Emilia-Romagna dealer inventory.
• compare_vehicles returns a contextual comparison, not just a list of fields.
• evaluate_trade_in prepares an indicative range with explicit assumptions.
• draft_test_drive_request prepares a request for human review.

The interface shows the same work to the person: the active brief, the shortlist, the comparison board, the trade-in context, and the final handoff. Search and comparison calls update that visible state directly, so the agent and the person stay in the same showroom context. Every action in the demo is draft-only: no message is sent, no appointment is booked, and no real lead is created.

How we built it

The public repository is a small Next.js application with a typed synthetic inventory, client-side WebMCP feature detection, and four structured tools registered through document.modelContext. The UI is intentionally usable without an agent, while capable browsers can discover the same actions as tools.

Why WebMCP matters here

WebMCP lets a website expose capabilities instead of forcing an agent to guess through clicks and visual labels. The Dealer's Desk uses that capability to make the first steps of a dealership journey more useful while keeping scope visible and human-controlled.

Built With

• next.js
• react
• typescript