The extraction API that shows its work. Send a URL or raw HTML, get structured JSON with per-field confidence scoring and extraction provenance — every field shows which method produced it (Schema.org, LLM inference, or headless browser) and how confident the system is. Set strict_confidence_threshold and uncertain fields are removed server-side before they reach your agent. 50 free calls/month.
Website: https://shopgraph.dev | API: https://shopgraph.dev/api/enrich/basic | MCP: https://shopgraph.dev/mcp
UCP output validated with ucp-schema v1.1.0 — the official Universal Commerce Protocol schema validator.
# Free — no API key, no signup
curl -X POST https://shopgraph.dev/api/enrich/basic \
-H "Content-Type: application/json" \
-d '{"url": "https://www.allbirds.com/products/mens-tree-runners"}'Returns structured JSON with per-field confidence scores:
{
"product": {
"product_name": "Men's Tree Runners",
"brand": "Allbirds",
"price": { "amount": 100, "currency": "USD" },
"availability": "in_stock",
"categories": ["Shoes", "Running"],
"confidence": { "overall": 0.95 },
"_shopgraph": {
"field_confidence": {
"product_name": 0.97,
"brand": 0.95,
"price": 0.98,
"availability": 0.90
}
}
},
"free_tier": { "used": 1, "limit": 50 }
}| Tool | REST Endpoint | Price | What It Does |
|---|---|---|---|
enrich_basic |
POST /api/enrich/basic |
Free (shared quota) | Schema.org extraction only. Fast, zero LLM cost. |
enrich_product |
POST /api/enrich |
Free 50/mo, then subscription or $0.02/call | Full pipeline with per-field confidence scoring and extraction provenance. |
enrich_html |
POST /api/enrich/html |
Subscription or $0.02/call | Bring your own HTML. Works with Bright Data, Firecrawl, or any fetch/proxy tool. |
Pricing: Free (50/mo) | Starter $99/mo (10K calls) | Growth $299/mo (50K calls) | Enterprise (custom). Pay-per-call via Stripe MPP still available for agents. Cached results (24h) are free. No charge for failed extractions.
Your agent sends a URL (or raw HTML)
→ Tier 1: Schema.org/JSON-LD parsing (0.93 baseline confidence, instant)
→ Tier 2: LLM extracts from page text when structured data is absent (0.70 baseline)
→ Tier 3: Headless Playwright renders JavaScript, then extracts (additional inference step)
→ Returns ProductData with per-field confidence scores and extraction provenance
(which tier produced each field) in _shopgraph.field_confidence
→ Set strict_confidence_threshold to remove low-confidence fields server-side
before they reach your agent
→ Add format=ucp for Universal Commerce Protocol output
Authentication: API key (sg_live_ keys) for subscription tiers, or Stripe MPP for pay-per-call agents.
ShopGraph is a structuring layer, not a fetcher. It's complementary to Bright Data, Firecrawl, and other fetch/proxy tools. They handle retrieval. ShopGraph handles extraction provenance and per-field confidence scoring.
curl -X POST https://shopgraph.dev/api/enrich/basic \
-H "Content-Type: application/json" \
-d '{"url": "https://example.com/product"}'Schema.org only. Shares the free-tier quota with /api/enrich. No signup needed.
# With API key (subscription)
curl -X POST https://shopgraph.dev/api/enrich \
-H "Content-Type: application/json" \
-H "Authorization: Bearer sg_live_..." \
-d '{"url": "https://example.com/product", "strict_confidence_threshold": 0.8, "format": "ucp"}'
# With Stripe MPP (pay-per-call)
curl -X POST https://shopgraph.dev/api/enrich \
-H "Content-Type: application/json" \
-d '{"url": "https://example.com/product", "payment_method_id": "pm_..."}'Full pipeline: Schema.org → LLM inference → headless browser. 50 free calls/month. Authenticate with API key (sg_live_) or Stripe MPP for higher limits.
curl -X POST https://shopgraph.dev/api/enrich/html \
-H "Content-Type: application/json" \
-d '{"html": "<html>...</html>", "url": "https://example.com/product", "payment_method_id": "pm_..."}'Already fetched the page? Pipe the HTML to ShopGraph for structuring.
{
"mcpServers": {
"shopgraph": {
"type": "url",
"url": "https://shopgraph.dev/mcp"
}
}
}Works with Claude, Claude Code, Cursor, Windsurf, CrewAI, LangGraph, AutoGen, and any MCP client.
Every response includes:
| Field | Description |
|---|---|
product_name |
Product title |
brand |
Manufacturer or brand |
price |
Amount + currency + sale price |
availability |
in_stock, out_of_stock, preorder, unknown |
categories |
Product taxonomy |
image_urls |
Product images (enrich_product/enrich_html only) |
color |
Available colors |
material |
Materials/fabrics |
dimensions |
Size/weight info |
confidence |
Overall + per-field scores (0-1) |
_shopgraph.field_confidence |
Per-field confidence with field-type modifiers |
git clone https://github.com/laundromatic/shopgraph.git
cd shopgraph
npm installRequired .env:
| Variable | Purpose |
|---|---|
STRIPE_TEST_SECRET_KEY |
Stripe secret key (test or live) |
GOOGLE_API_KEY |
Gemini API key for Tier 2 (LLM) inference |
UPSTASH_REDIS_REST_URL |
Upstash Redis for stats/monitoring (optional) |
UPSTASH_REDIS_REST_TOKEN |
Upstash Redis token (optional) |
npm run build # Compile TypeScript
npm start # Run MCP server (stdio)
npm run start:http # Run HTTP server
npm run dev # Dev mode (no build needed)
npm run test:run # Run 118 testsShopGraph runs 118 automated tests across 22 product verticals. Self-healing pipeline with circuit breaker, URL verification, and health alerts.
- Health: https://shopgraph.dev/health
- Stats: https://shopgraph.dev/api/stats
- Dashboard: Live on shopgraph.dev homepage
Apache 2.0
Krishna Brown | Los Angeles, CA