Technical guide
Build a classic car valuation agent with MCP
This walkthrough shows how to connect an AI agent to Old Cars Data so it can normalize a collector car, find comparable sales, check live auction activity, inspect bid trails, and summarize market statistics without scraping auction websites.
Last updated May 18, 2026
What the agent should do
A useful valuation agent should not return a single unsupported number. It should gather comparable sales, separate completed outcomes from live auction signals, summarize price statistics, and explain where the estimate is weak.
| Step | MCP tool | Purpose |
|---|---|---|
| Normalize the vehicle | list_makes, list_models | Use canonical names before searching. |
| Find comparable sales | search_auctions | Return completed listings, prices, statuses, and vehicle details. |
| Check the live market | search_live_auctions | Compare current bidding against completed results. |
| Inspect demand | get_auction_bids | Review bid progression where supported. |
| Summarize the market | get_price_stats | Add median, range, trend, and volume context. |
Connect the MCP server
Compatible clients can connect to the remote MCP endpoint at https://api.oldcarsdata.com/mcp. OAuth handles authentication in clients that support it. Developer tools that use project config can add the server like this:
{
"mcpServers": {
"oldcarsdata": {
"type": "http",
"url": "https://api.oldcarsdata.com/mcp"
}
}
}See the full MCP setup guide for ChatGPT, Claude, Cursor, Windsurf, and Claude Code.
Give the agent a valuation policy
The prompt should force the agent to use normalized data, cite the market signals it checked, and avoid overstating confidence.
You are a collector-car market research assistant.
When a user asks what a car is worth:
1. Use list_makes and list_models to normalize the make and model.
2. Use search_auctions to find comparable completed sales.
3. Use search_live_auctions to check current market activity when relevant.
4. Use get_auction_bids for supported listings when bid behavior matters.
5. Use get_price_stats for market-level context.
6. Explain confidence, caveats, and what a human should inspect next.
Do not call the result a formal appraisal.Example user request
A user can ask: “What is a fair bid limit for a 1997 Toyota Land Cruiser?” The agent should normalize Toyota and Land Cruiser, retrieve recent completed auctions, check live listings if the user is evaluating an active auction, pull bid history where available, and return a range with caveats.
Good output shape
- Comparable sales reviewed, including dates and outcomes.
- Median and range context from market statistics.
- Live auction differences such as mileage, color, condition, and bid count.
- Confidence level and missing inspection details.
- Clear disclaimer that this is market research, not a formal appraisal.
Frequently asked questions
What is a collector-car valuation agent?
A collector-car valuation agent is an AI workflow that can identify a vehicle, query comparable auction results, inspect live market activity, and summarize pricing context for a human reviewer.
Does the agent produce a formal appraisal?
No. The workflow produces market research and comparable-sale context. Formal appraisals should include condition, provenance, inspection, title history, documentation, and expert judgment.
Why use MCP for this workflow?
MCP lets the AI client call structured tools against current collector-car data instead of relying on stale model memory, search snippets, or scraping auction sites.