ShopGraph
A product data structuring layer for agent commerce.
ShopGraph turns raw product pages into clean, confidence-scored, agent-ready JSON. It works alongside scraping tools like Bright Data and Firecrawl, not instead of them. They get the HTML. ShopGraph makes it useful.
Extraction follows three tiers. Schema.org and JSON-LD parsing runs first: fast, free to compute, 0.95 confidence. When structured markup is missing, a Gemini LLM reads the page content and extracts product attributes. For JavaScript-rendered or bot-protected pages, a Playwright browser renders the page before extraction. Every field in the response includes a confidence score so agents know how much to trust it.
Reliability monitoring is not an afterthought. ShopGraph tests 117 verified URLs across 22 verticals every 30 minutes. A circuit breaker auto-quarantines URLs after consecutive failures. A verification gate checks URLs before they enter the test corpus. A health endpoint reports degradation in real time. The monitoring infrastructure is as much work as the extraction engine. For infrastructure that agents call programmatically, that is the right ratio.
POST /api/enrich/basicPOST /api/enrichPOST /api/enrich/htmlREST API at shopgraph.dev/api. MCP protocol at shopgraph.dev/mcp. 200 free calls per month, no signup. Paid calls via Stripe Machine Payments Protocol. Apache 2.0 on GitHub.
Where It Fits
From a URL
POST /api/enrich with a product URL. ShopGraph fetches, parses, and structures. Schema.org first, LLM when markup is missing, browser rendering for JS-heavy pages. For when you don't have the HTML yet.
From HTML
POST /api/enrich/html with raw markup your scraper already fetched. ShopGraph structures it without re-fetching. Works with Bright Data, Firecrawl, Playwright, or any scraping tool. One schema out, regardless of what went in.
From any agent
Connect via MCP protocol at shopgraph.dev/mcp. Same extraction engine, native agent integration. Works with Claude, Cursor, CrewAI, LangGraph, AutoGen, and any MCP client.
How ShopGraph Evolved
Shipped as an MCP server extracting structured product data from any URL. First MCP server for open-web product data. Stripe Machine Payments Protocol integration from day one.
Published to Glama (AAA quality rating), Smithery, and mcpservers.org. Self-healing test pipeline operational: circuit breaker, URL verification gate, health monitoring.
Stopped competing with scraping infrastructure. Added enrich_html: agents bring their own HTML from any scraping tool, ShopGraph structures it. Complementary to Bright Data, Firecrawl, and Apify rather than competing with them.
Moved beyond MCP-only distribution. REST API makes ShopGraph accessible from any language, any framework. Free tier removes the evaluation barrier: 200 calls per month, no signup needed.
Live playground on the homepage — paste a URL, get structured data back, no signup. npm client SDK for JavaScript/TypeScript developers. Three integration paths: REST API, MCP protocol, npm SDK.
Wiring SceneInBloom's Product Intelligence Agent as the first production consumer. Expanding to additional MCP directories and developer marketplaces.
Where ShopGraph Came From
ShopGraph originated inside SceneInBloom, a platform where AI agents curate shoppable lifestyle content from cinematic moments. SceneInBloom's multi-agent infrastructure — workflow coordination, editorial voice systems, product intelligence pipelines — is producing standalone tools. ShopGraph was the first.
Built by Krishna Brown
Krishna builds agent orchestration systems and human-AI collaboration tools — working with AI to ship production infrastructure. Her focus is workflow coordination, multi-agent systems, agentic commerce and payments, and making the handoff between humans and AI agents actually work.
hi@kb.computer · GitHub · Apache 2.0