CXH AUTO: querying China car-price trends through MCP

Description and use

CXH AUTO delivers precise car prices in China by preserving price categories, handling sparse city data, and verifying membership after login.

Problem

Ensuring that price answers match the buyer’s actual question when querying China new‑ and used‑car market prices via a remote MCP connection.

Outcome

The beta Skill now returns price data with explicit categories, preserves query context, and verifies membership, reducing ambiguous answers and improving developer feedback loops.

Workflow

• 1. Agent requests price for a specific vehicle configuration. 2. Skill queries MCP endpoint with configuration, region, and time. 3. MCP returns observations with price category and conditions. 4. Skill filters results: exact configuration first, then same‑series alternatives. 5. Skill presents comparison view, keeping original used‑car query intact. 6. If login succeeded, Skill verifies membership state before proceeding. 7. Agent receives categorized price data and can choose trim or alternative models.

Evidence

The interesting engineering problem is not connecting a language model to a search endpoint. It is making sure that the number returned answers the buyer's actual question.

Source

Our rule is to preserve the price category supplied by the observation.

Source

If the requested city has too few observations, a provincial reference can still be useful.

Source

One failure mode during testing was an apparently completed login while the original agent connection still behaved as a guest.

Source