The extraction API that
shows its work.
ShopGraph extracts product data from any URL and shows you exactly how each field was derived. Per-field provenance and confidence based on extraction method, not opinion. Your automation decides what's reliable enough to act on.
Confidence Scoring
Every extracted field includes a confidence score from 0.0 to 1.0 showing how it was derived. Set a threshold (e.g., 0.85) and the engine filters fields that fall below it. Your agents act on data with visible provenance — every field shows how it was derived and which extraction method produced it.
Native UCP Output
Standardize your entire catalog pipeline. ShopGraph extracts complex variant matrices and instantly normalizes them into Universal Commerce Protocol JSON, strictly validated against the official ucp-schema. One schema for every merchant.
AgentReady Scoring
Score how well a site supports automated extraction before committing pipeline resources. ShopGraph evaluates target URLs across five active dimensions of machine-readability and returns a 0-100 score.
Self-Healing Fallbacks
Three extraction methods cross-check each other. When two tiers agree on a price and one dissents, the confidence score reflects that signal. The pipeline: Schema.org → LLM extraction → Headless Playwright rendering. Self-healing means the confidence contract is calibrated, not just the extraction.
RFC 9421 Signature Compliance
ShopGraph signs outbound requests with RFC 9421 (Ed25519). Commerce sites in the Cloudflare Signed Agents directory can verify ShopGraph as a registered agent, enabling fast-lane access past bot detection. This does not bypass account authentication or commercial agreements.
Design Principles
Every field tells you how confident the system is. You decide what to trust.
ShopGraph handles the extraction pipeline. Your agents get structured data with confidence scores. You don't manage proxies or extraction infrastructure.
Every response follows the same structure regardless of the source site. Write one parser, use it everywhere.
Every response shows which extraction method produced each field. Your downstream systems make informed decisions.
All data comes back with quality signals attached. You set the bar for what's trusted, flagged, or rejected.