What it is
SEO intent-fit scoring with a calibrated noise band.
What it does: Judges term presence and overall intent fit with typed Jev Score questions, reports a noise floor from repeated runs, and exposes the same engine over MCP for agents.
Why it belongs in the showcase: A shipped product loop (web + MCP) that turns Jev ordinal scores into an inspectable page-quality signal with explicit measurement limits.
Caveats: Not a ranking predictor; key required for live Jev.
Evidence
Public MIT repo with a live demo at https://seo-jev-analyzer.vercel.app. Each page gets a 0–10 score = (sum of ten term ratings) × intent fit; each rating is a five-level Jev Score whose value is the probability-weighted average level. Metadata and full-page scores are reported separately. A consistency report on the app scores 10 public pages 100 times each to set the noise floor. Browser data stays in IndexedDB; the TypeSafe key stays on the server. An MCP endpoint exposes score_page and related tools.
Caveats
- 0★ at intake
- Scores reflect Jev relevance judgment, not search ranking or AI-answer visibility
- Needs a TypeSafe API key on the server for real scores
Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.