carsensor Price Range MCP Server for AI Agents

If you're building an AI agent, chatbot, or app that needs to answer "is this used car's asking price fair?" — not a guess, but a number grounded in real market data — this MCP server gives you exactly that in a single tool call. resolvePriceRange(carModel, year) returns a market-price range (min / max / median) computed live from carsensor.net's own official price-by-model-year market statistics — the same data carsensor itself publishes on its 相場(market price) pages, turned into a structured, ready-to-use number instead of a page you'd have to read and interpret yourself.

What you get:

• One tool, resolvePriceRange, taking just a car model name and a year — no API keys to manage for the underlying data source, no HTML to parse yourself
• priceRangeYen: { min, max, median } plus sampleSize (how many real carsensor listings the estimate is based on) and confidence (normal / low — honestly downgraded, not hidden, when the year can't be fully verified against carsensor's own data; see "Known limitations")
• Always-on hosting via Apify Standby mode — no server to run yourself. Connect any MCP client (Claude Desktop, Claude Code, your own agent) directly to a stable HTTPS endpoint, or call it as a plain Streamable HTTP MCP server from your own app's code

{
  "ok": true,
  "carModel": "プリウス",
  "year": 2018,
  "priceRangeYen": { "min": 750000, "max": 4772500, "median": 1573112 },
  "sampleSize": 412,
  "confidence": "normal",
  "disclaimer": "carsensor.net上の公開情報(相場ページの価格×年式集計)を基にした参考値です。個体差・装備差・実際の商談結果は反映されません。"
}

Pay-per-event pricing — see "Pricing" below. Full input/output details, setup, and technical design are further down this page.

⚠️ Disclaimer (please read)

• This Actor automatically retrieves information that is publicly available on carsensor.net. You are responsible for how you retrieve and use this data.
• The site owner's Terms of Use, robots.txt, or access blocking could change without notice, which could stop this Actor from working or change its output unexpectedly.
• priceRangeYen is a statistical estimate derived from carsensor's own published price×model-year aggregate data, not a guarantee of any individual vehicle's true value. It does not account for condition, trim/grade differences, or mileage (see "Known limitations"). Always verify a specific vehicle's price on carsensor.net itself before making any purchase or sale decision based on this data.

Who is this for?

• AI agents and chatbots that need to ground a "what is this actually worth?" answer in real market data instead of the model's own (often outdated or hallucinated) knowledge of used-car pricing
• App/tool developers building car-shopping, valuation, or trade-in estimation features who want a ready-made price-range lookup instead of scraping and maintaining their own carsensor.net integration
• Anyone who wants this as a plain MCP tool call rather than a batch scrape — this Actor is the single-tool, call-on-demand counterpart to our batch-oriented carsensor-resale-value-scout Actor (which scores whole lists of listings against the same underlying market data)

What can't it do (please read before integrating)

• Mileage is not a supported input yet. Only car model and year are used — the price range reflects carsensor's own model-year aggregate, not a mileage-adjusted estimate. A future version may add this.
• Model-name resolution can occasionally fail or be slow on first use for an unfamiliar model — see "Known limitations" for why, and how caching mitigates this after the first call.
• This is a statistical estimate, not an appraisal. See the Disclaimer above.

Input / Output

Input (MCP tool call arguments):

Output (success):

Output (model not found / no data for that year): { "ok": false, "reason": "no_code_resolved" | "no_matching_data", "message": "..." } — not charged (see "Pricing").

Usage example

Claude Desktop / any MCP client config:

{
  "mcpServers": {
    "carsensor-price-range": {
      "url": "https://woolen-snake--carsensor-price-range-mcp.apify.actor/mcp",
      "headers": { "Authorization": "Bearer YOUR_APIFY_TOKEN" }
    }
  }
}

Calling it directly from your own app (Node.js, @modelcontextprotocol/sdk):

import { Client } from '@modelcontextprotocol/sdk/client/index.js';
import { StreamableHTTPClientTransport } from '@modelcontextprotocol/sdk/client/streamableHttp.js';

const transport = new StreamableHTTPClientTransport(
  new URL('https://woolen-snake--carsensor-price-range-mcp.apify.actor/mcp'),
  { requestInit: { headers: { Authorization: `Bearer ${process.env.APIFY_TOKEN}` } } },
);
const client = new Client({ name: 'my-app', version: '1.0.0' });
await client.connect(transport);

const result = await client.callTool({
  name: 'resolvePriceRange',
  arguments: { carModel: 'プリウス', year: 2018 },
});
console.log(JSON.parse(result.content[0].text));

Pricing

Flat pricing regardless of whether the call hit cached data or triggered a fresh request to carsensor.net — see "How the price was set" below for the full reasoning, including why cache-tiered pricing was deliberately rejected.

How the price was set

$0.03 sits between this Actor family's base classification tier ($0.015, e.g. listing-extracted on our carsensor-resale-value-scout Actor) and its heavier derived-signal tier ($0.045–$0.065, e.g. value-score-computed) — reflecting that a price range is a computed market signal, not a raw data fetch, while this tool's underlying infra cost (well under $0.001/call, most calls served from cache) leaves a large margin either way.

Why cache hits and misses are priced the same: the value delivered — priceRangeYen, sampleSize, confidence — is identical either way; caching is purely an internal performance/cost optimization, not a difference in what you get. Pricing it differently would also create a perverse incentive to shorten the cache TTL (more misses = more revenue), which runs directly against this tool's own design goal of minimizing load on carsensor.net.

Data source and calculation

Uses carsensor.net's own publicly-published 相場(market price) page for the resolved model (/usedcar/souba/{makerCode}_S{modelCode}/), specifically its "price × model-year" cross-tabulation (price bucket, year bucket, listing count per cell) — official aggregate data carsensor itself computes from its own live inventory, not a third-party estimate.

• median is a count-weighted average of each matching year-bucket's price-bucket midpoint
• min / max come from the actual price-bucket bounds of the matching cells (open-ended buckets, e.g. "¥4,150,000 or more", are approximated with a ±15% heuristic — an empirical adjustment, not a statistically derived one)
• This is grade-level aggregate data (e.g. all Toyota Prius trims combined), not individual-listing pricing — it doesn't account for trim/grade differences, equipment, or condition

Caching (2-tier)

A cache hit on both tiers means zero HTTP requests to carsensor.net for that call. On the Apify-hosted version, this cache is scoped to a single Standby run/container instance — concurrent sessions can land on different container instances with independent caches, so the real-world hit rate on the hosted version may be lower than what a single long-running local process would see. A future version may move this to Apify's Key-Value Store (shared across runs) if this turns out to matter in practice.

Access interval and logging

Requests to carsensor.net are paced at a minimum 1000ms interval (1500ms by default) as a load-reduction measure — the same floor used by our carsensor-resale-value-scout Actor. Every call is logged (model, year, per-tier cache hit/miss, HTTP requests actually made, response time, outcome) for future monitoring — not exposed to callers, used only for operating this Actor responsibly.

FAQ

Why is confidence sometimes "low"? carsensor's year axis has an open-ended bottom bucket in its own data (e.g. "2012 or earlier"). Any year at or before that bound will match it, whether or not the model actually existed yet in that year — there's no way to fully verify plausibility from this data alone, so confidence is honestly downgraded rather than reported as certain. See the note field on such responses for the specific reason.