Nuke MCP Server

Vehicle intelligence for AI agents. Search 1.29M+ vehicle profiles, get comparable sales, AI valuations, $0/image analysis, and listing extraction.

npx -y @sss97133/nuke-mcp-server

What it does

Nuke is a vehicle data platform with 1.29M profiles, 1.36M observations, 11.7M auction comments, and 35M images aggregated from Bring a Trailer, Cars & Bids, RM Sotheby's, Mecum, Barrett-Jackson, eBay Motors, Craigslist, forums, and 100+ more sources.

This MCP server gives any AI agent access to that data through 12 tools:

| Tool | Description | |------|-------------| | search_vehicles | Quick search by VIN, URL, year, make/model, or free text | | search_vehicles_api | Full-text search with filters (make, model, year, price), pagination, and inline valuations | | extract_listing | Extract structured data from any car listing URL | | get_vehicle_valuation | Compute multi-signal market valuation ("The Nuke Estimate") | | get_valuation | Look up cached valuation by vehicle_id or VIN | | identify_vehicle_image | AI vision: photo -> year/make/model/trim with confidence | | analyze_image | YONO vision: make, condition, zone, damage, mods ($0/image) | | get_comps | Comparable auction sales with price statistics | | get_vehicle | Fetch a vehicle profile by ID | | list_vehicles | List vehicles with filters (make, model, year, price range) | | ingest_marketplace_listing | Submit pre-extracted FB Marketplace data directly to Nuke | | import_facebook_saved | Import vehicles from Facebook Saved Items (browser extraction) |

Setup

1. Get an API key

Sign up at nuke.ag and generate an API key in Settings.

2. Configure your MCP client

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "nuke": {
      "command": "npx",
      "args": ["-y", "@sss97133/nuke-mcp-server"],
      "env": {
        "NUKE_API_KEY": "nk_live_your_key_here"
      }
    }
  }
}

Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "nuke": {
      "command": "npx",
      "args": ["-y", "@sss97133/nuke-mcp-server"],
      "env": {
        "NUKE_API_KEY": "nk_live_your_key_here"
      }
    }
  }
}

Cursor

Add to Cursor Settings -> MCP Servers:

{
  "nuke": {
    "command": "npx",
    "args": ["-y", "@sss97133/nuke-mcp-server"],
    "env": {
      "NUKE_API_KEY": "nk_live_your_key_here"
    }
  }
}

Example prompts

Once configured, you can ask your AI agent:

• "What's a 1973 Porsche 911 worth right now?"
• "Find comparable sales for a 1967 Ford Mustang Fastback"
• "Extract the vehicle data from this BaT listing: https://bringatrailer.com/listing/..."
• "What car is in this photo?" (with image URL)
• "Analyze this car photo for condition and damage" (with image URL)
• "Search for all BMW M3 E30s in the database"
• "Look up VIN WP0AB2966NS420176"

Tools in detail

search_vehicles

Quick search. Accepts any input and auto-detects query type.

Input:  { query: "1967 Mustang", limit: 5 }
Output: { results: [...], query_type: "text", total_count: 142 }

Query types auto-detected:

• VIN (17 chars) -> direct vehicle lookup
• URL -> routes to extraction
• Year (4 digits) -> vehicles from that year
• Text -> multi-entity search across vehicles, orgs, users

search_vehicles_api

Full-text search with filters, pagination, sort, and inline valuations.

Input:  { q: "Porsche 911", make: "Porsche", year_from: 1990, year_to: 2000, limit: 50 }
Output: { data: [...], pagination: { page: 1, total_count: 312 }, search_time_ms: 45 }

Each result includes valuation with estimated_value and confidence_score when available.

extract_listing

Works on any car listing URL. No configuration needed.

Input:  { url: "https://bringatrailer.com/listing/1988-porsche-911-turbo-68" }
Output: { year: 1988, make: "Porsche", model: "911", trim: "Turbo", vin: "...", ... }

Supported sources include: Bring a Trailer, Cars & Bids, Hagerty, PCarMarket, RM Sotheby's, Mecum, Gooding, Bonhams, eBay Motors, Craigslist, Facebook Marketplace, Collecting Cars, and any generic listing page.

get_vehicle_valuation

Computes the Nuke Estimate using 8 signals:

1. Comparable sales (recency-weighted)
2. Condition assessment
3. Rarity (production data + survival rates)
4. Sentiment (auction comment analysis)
5. Bid curve velocity
6. Market trend (30d/90d)
7. Originality score
8. Record proximity

Input:  { vehicle_id: "uuid" }
Output: { estimated_value: 47500, value_low: 41000, value_high: 54000,
          confidence_score: 0.82, deal_score: "Good Deal", heat_score: 0.7,
          price_tier: "mainstream" }

get_valuation

Fast cached valuation lookup (no recomputation).

Input:  { vehicle_id: "uuid" }  or  { vin: "WP0AB2966NS420176" }
Output: { data: { estimated_value: 47500, confidence_score: 0.82, deal_score: "buy", ... } }

identify_vehicle_image

Tiered AI approach: Gemini Flash -> GPT-4o-mini -> GPT-4o for cost efficiency.

Input:  { image_url: "https://example.com/car.jpg" }
Output: { year: 1973, make: "Porsche", model: "911", trim: "Carrera RS",
          body_style: "coupe", generation: "G-Series", confidence: 0.87,
          reasoning: "Distinctive ducktail spoiler, Carrera side script..." }

analyze_image

YONO-powered deep analysis. $0/image (local inference, zero cloud API calls).

Input:  { image_url: "https://example.com/car.jpg", include_comps: true }
Output: { make: "Porsche", confidence: 0.91, family: "german",
          vehicle_zone: "ext_front_driver", condition_score: 4,
          damage_flags: [], modification_flags: ["aftermarket_wheels"],
          photo_quality: 4, source: "yono", cost_usd: 0,
          comps: [...] }

get_comps

Comparable vehicle sales from real auctions.

Input:  { make: "Porsche", model: "911", year: 1973, limit: 10 }
Output: { data: [...], summary: { count: 10, avg_price: 142000, median_price: 135000,
          min_price: 89000, max_price: 245000, auction_event_count: 8 } }

Data from: Bring a Trailer, Mecum, Barrett-Jackson, RM Sotheby's, Cars & Bids, Gooding, Bonhams, PCarMarket, eBay Motors, and more.

ingest_marketplace_listing

Submit pre-extracted Facebook Marketplace vehicle data directly to Nuke. Use this when you've already scraped data from a FB Marketplace listing page (e.g., via DOM extraction in a browser). Skips re-scraping — data goes straight to the database. Idempotent on facebook_id.

Input:  { facebook_id: "1234567890", title: "1984 Chevrolet K10",
          price: 18500, parsed_year: 1984, parsed_make: "Chevrolet",
          parsed_model: "K10", location: "Austin, TX",
          mileage: 87000, all_images: ["https://..."] }
Output: { success: true, listing_id: "uuid", vehicle_id: "uuid" | null,
          is_new: true, submission_count: 1, linked_via: "vin" | "ymm_match" | null }

Automatic vehicle linking: if the description contains a VIN matching an existing vehicle, vehicle_id is set directly. If year/make/model + state match 1-3 existing vehicles, suggested_vehicle_id is set for human review.

import_facebook_saved

Import vehicles from a user's Facebook Saved Items. Two-step flow using browser-side extraction.

Step 1: Get the extractor script

Input:  { action: "extract" }
Output: JavaScript snippet to execute on facebook.com/saved

Step 2: Submit extracted vehicles

Input:  { action: "submit", vehicles: [{ title: "1976 Pontiac Bonneville",
          year: 1976, price: 6000, seller: "Jane Doe", sold: false,
          facebook_id: "1987905388825064",
          url: "https://www.facebook.com/marketplace/item/1987905388825064/",
          parsed_make: "Pontiac", parsed_model: "Bonneville" }, ...] }
Output: { total_submitted: 485, inserted: 460, skipped: 20, errors: 5,
          sold_count: 323, active_count: 162 }

Agent workflow:

1. Call import_facebook_saved(action: "extract") to get the browser JS
2. Navigate user's browser to facebook.com/saved
3. Execute the JS snippet (auto-scrolls and extracts all saved marketplace listings)
4. Read window.__nuke_fb_saved.vehicles from the browser
5. Call import_facebook_saved(action: "submit", vehicles: [...]) to ingest into Nuke

get_vehicle / list_vehicles

Standard CRUD for vehicle profiles with full filter support.

Environment variables

| Variable | Required | Description | |----------|----------|-------------| | NUKE_API_KEY | Yes* | Your Nuke API key (nk_live_...) | | NUKE_SERVICE_ROLE_KEY | No | Supabase service role key (internal/dev use) | | NUKE_API_URL | No | API base URL (defaults to production) |

*Either NUKE_API_KEY or NUKE_SERVICE_ROLE_KEY is required.

Data

• 1,290,000+ vehicle profiles
• 1,360,000+ observations with full provenance
• 35M+ images indexed
• 11.7M+ auction comments analyzed
• 4.1M+ bids tracked
• 773K+ valuations at 6.3% MAPE
• 3,987 businesses identified
• 112+ data sources

All data points have provenance tracking and confidence scores. BaT has NO public API -- Nuke is the only way to access structured BaT data programmatically.

Built by

Solo founder + Claude. The entire platform -- 300+ edge functions, 1000+ database tables, React frontend, Elixir API, YONO vision model -- was built by one person with AI.

License

MIT