Loading ToolBench
MCP server for Nuke vehicle data platform — 1.29M vehicle profiles, AI valuations, $0/image vision analysis, comparable sales, listing extraction
This server demonstrates good definition quality with consistent naming conventions, comprehensive descriptions, and well-structured input schemas. All 11 tools follow verb_noun patterns (search_, extract_, get_, list_, identify_, analyze_, ingest_). Descriptions are detailed and contextually helpful (averaging ~180 chars), explaining WHAT each tool does and WHEN to use it. Input schemas use Zod with proper types and constraints. However, there are notable gaps: no documented output schemas for any tool, limited error handling guidance in descriptions, and no tool annotations (readOnlyHint, destructiveHint, idempotentHint). The server correctly marks all tools as READ_ONLY in risk assessment but does not expose this via MCP tool annotations. Parameter descriptions are generally strong (averaging ~70 chars) with good use of constraints (enums, min/max, format). Most tools accept flexible human-friendly identifiers (e.g., vehicle_id OR vin+year+make+model in get_vehicle_valuation) which matches the chat data model well.
YONO vision: $0/image analysis with make, condition, zone, damage, mods. Uses Vision API to extract: car make, overall condition score, damaged zones (e.g., 'front-left'), damage types (e.g., 'paint', 'dent'), and modifications (e.g., 'lowered suspension'). Perfect for valuation adjustment and condition assessment.
Extract structured vehicle data from ANY car listing URL. Works on Bring a Trailer, Cars & Bids, Craigslist, eBay Motors, Facebook Marketplace, Hagerty, PCarMarket, RM Sotheby's, Mecum, and thousands of other sites. Returns year, make, model, VIN, price, mileage, images, and more.
Comparable vehicle sales by make/model or vehicle_id. Returns recent sales history with price, mileage, condition, and date sold. Useful for market analysis and valuation validation.
Cached Nuke Estimate lookup by vehicle_id or VIN. Returns the last computed valuation without recomputing. Faster and cheaper than get_vehicle_valuation for bulk lookups. Use get_vehicle_valuation to force a fresh computation.
Fetch a vehicle profile by ID. Returns comprehensive vehicle data: year, make, model, VIN, auction history, photos, estimated value, and more.
No documented output schemas for any of the 11 tools. LLMs cannot plan downstream tool calls or extract needed fields from responses without knowing what data structures to expect.
No tool annotations in MCP schema. All tools are READ_ONLY (documented in risk assessment) but this is not exposed via MCP readOnlyHint. Agents cannot distinguish read-only tools from state-modifying ones without this annotation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 58 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 17 | - | v1 |
Compute 'The Nuke Estimate' -- a confidence-weighted multi-signal valuation. Uses 8 signals: comparable sales, condition, rarity, sentiment, bid curves, market trends, survival rates, and originality. Returns estimated value, confidence score, deal score, heat score, and price tier (budget/mainstream/enthusiast/collector/trophy). For cached valuations without recompute, use get_valuation instead.
AI vision: upload a photo and get year/make/model/trim identification. Sends image to Claude 3.5 Sonnet for analysis. Useful for identifying unknown vehicles or validating seller claims.
Direct FB Marketplace listing ingest (no re-scrape). Provide a Facebook Marketplace URL and Nuke will extract and parse the listing data without re-fetching.
List vehicles (yours or public). Returns paginated list of vehicle profiles with pagination support.
Quick search the Nuke database (1.29M+ vehicle profiles). Accepts any input: VIN (17 chars), listing URL, year, make/model text, or free-text query. Returns matching vehicles with thumbnails. For filtered search with pagination and valuations, use search_vehicles_api instead.
Full-text search across 1.29M+ vehicles with filters, pagination, and inline valuation data. Filter by make, model, year range. Results include VIN, price, mileage, color, transmission, body style, auction source, and Nuke Estimate valuation when available. Sorted by relevance, price, or year.
Limited error handling guidance. Descriptions do not include 'what to do if this fails' guidance. For example, search_vehicles does not explain what happens if no results are found or how to retry with different query terms.
identify_vehicle_image and analyze_image descriptions mention sending to Claude 3.5 Sonnet or Vision API but do not document rate limits, retry behavior, timeout handling, or cost implications for agents making bulk calls.
list_vehicles and get_comps accept optional pagination parameters (limit, page, default values) but description does not explain total count or next_cursor behavior, making it unclear if results are complete or truncated.
extract_listing tool description mentions many supported platforms (Bring a Trailer, Cars & Bids, Craigslist, eBay Motors, etc.) but no guidance on what happens for unsupported platforms or partial extraction failures.