scentrev-mcp logoscentrev-mcp

Changelog

Perfume MCP changelog

Release notes for the scentrev-mcp perfume MCP server and fragrance API: new tools, reliability fixes, and docs updates.

Fixedv3.4.3August 20, 2026

MCP-UI actually embeds — HTML resources + nested profile visuals

  • get_fragrance_profile now attaches pyramid / season / radar from nested sections (note_pyramid, accords.accords, performance.season), not only top-level keys.
  • UI blocks are HTML (text/html;profile=mcp-app) on the tool-result content array so MCP-UI / MCP Apps hosts can iframe them. ui://scentrev/visuals is registered as the shared app shell.
  • Cursor chat iframes that shell via the MCP Apps App Bridge (ext-apps ontoolresult + unpkg CSP), not a homemade postMessage listener.
  • Wear and season widgets show category labels (Long lasting, Excellent) instead of raw 0–1 scores or opaque percentages.
Newv3.4.2August 20, 2026

Fragrance MCP, perfume API, and ranking guides

  • New public pages for fragrance MCP, perfume API, and how-to guides. /fragrance-mcp is a full ranking pillar (FAQ, HowTo, tools) with the title Fragrance MCP.
  • Sitemap, llms.txt, HTML sitemap, and IndexNow now share one route list so every ranking URL is crawlable the day it ships.
Fixedv3.4.1August 18, 2026

Remaining audit gaps — fail-fast brands, blend pagination, MCP-UI, dupes

  • Unknown brand slugs now fail in search_fragrances_filtered with unknown_brand instead of timing out.
  • search_by_references scores only the active note_mode patterns, paginates past the first 10 rows, and uses the stored Bayesian rating column.
  • slug_aliases backfilled (~105k) and resolve_slug maps folded names to canonical slugs; unknown-brand sentinel rows dropped from ~1,836 to ~204 true unbranded.
  • get_accords score is 0–1; get_reminds_of includes n_records. Backend adds next_cursor, partial timeout retries, WWW-Authenticate on 401, a 20-connection HTTP pool, MCP-UI JSON blocks (pyramid / season wheel / radar), and find_dupes that respects max_price_ratio.
  • These backend/MCP-HTTP changes require a scentrev-mcp-backend deploy. Until then production MCP at api.scentrev.com is still the 14-tool build.
Fixedv3.4.0August 18, 2026

Audit PDF full implementation — correctness, performance, differentiation

  • P0-1/P0-2: split IdentityRow/ScoredRow/BlendRow schemas; slug_contains in resolve_slug enum; search_fragrances unblocked.
  • P1 fixes: unresolved[] on compare/blend, sections=[] invalid_param, identifier conflict guard, unified nested error envelope.
  • Performance migration: note_id index, name/brand trigram indexes, precomputed Bayesian rating, list_brands excludes unknown by default.
  • Token diet: $defs/$ref schemas, hoisted scale legends, verbosity=compact default on search tools.
  • OAuth discovery: /.well-known/oauth-protected-resource + authorization-server; rate-limit headers on 429.
  • Differentiation: 8 slash prompts, scentrev:// resources, find_dupes/blind_buy_risk/decode_taste, wardrobe tools + MCP-UI blocks.
Fixedv1.6.0August 4, 2026

Discovery hardening — timeouts, validation, ranking, pagination

  • Invalid category enums now return invalid_param immediately instead of rpc_timeout.
  • resolve_slug short-circuits on exact slug hits; no more full-catalog brand fan-out on informal names.
  • Default min_rating_votes raised from 20 → 200 (pass 0 to disable); timeout hints updated.
  • result_offset pagination on search, filtered search, resolve_slug, and list_brands.
  • Bayesian rating shrinkage + novelty-brand exclusion so meme entries stop winning serious sorts.
  • Rating categories recalibrated (excellent ≈ 4.3★+); reliability always present; price_value scale unified.
  • search_by_references samples notes across top/middle/base instead of top-only.
  • applied_defaults + explain=true for debugging; ignored_sections on profile; clean no_results errors.
Newv1.5.0August 3, 2026

Playground, recipes, and activation loop

  • Authenticated Playground runs all 14 MCP tools in the dashboard — first success without leaving the browser.
  • Onboarding checklist, canned recipes, and public /recipes + /claude-perfume-mcp pages for acquisition.
  • Up to two active API keys (dev/prod), soft monthly usage card, welcome backfill + day-3/7 activation drip.
  • Status page now probes the fragrance catalog; public_registry RLS enabled.
Improvedv1.4.1August 2, 2026

Weekly usage emails only for active API callers

  • Weekly digest emails now go only to users with more than one API call in that week — inactive accounts are skipped to cut inbox noise.
Fixedv1.4.0August 2, 2026

MCP search reliability — resolve_slug, matched notes, vote floors

  • Fixed resolve_slug so informal names like “le beau” resolve again (short brand false-matches and pooler-unsafe temp tables removed).
  • Restored matched_notes / matched_accords on search_fragrances_filtered (they were always empty after the top-10 fast path).
  • min_rating_votes now defaults to 20 in the database and MCP server when omitted (pass 0 to disable) so season browse no longer surfaces 1–2 vote perfect scores.
  • All 14 MCP tools re-verified end-to-end against live Supabase RPCs.
Improvedv1.3.0June 12, 2026

14-tool MCP surface and filtered search hardening

  • Expanded to 14 tools: list_brands, resolve_slug, compare_fragrances, plus blend via search_by_references.
  • Season browse injects min_season_category=good by default to avoid timeouts.
  • Normalized MCP errors (query_too_short, rpc_timeout, identifier_required) and ratio-safe rating payloads (no rating_avg).
  • Result hard-cap of 10 rows with truncated metadata on all list/search tools.
Newv1.2.0May 29, 2026

Status page, changelog, and richer docs

  • Live status page that checks the API and auth in real time.
  • REST examples (curl, JavaScript, Python) added to every tool in the docs.
  • Your active API key is now injected into copyable config when signed in.
Improvedv1.1.0May 22, 2026

Dashboard analytics overhaul

  • Replaced soft metrics with success rate, p95 latency, and error rate.
  • Per-request drill-down and CSV export on the usage log.
  • Rate-limit usage now visible at a glance.
Newv1.0.0May 15, 2026

scentrev-mcp is live

  • 10 MCP tools mapped 1:1 to the fragrance scoring model.
  • MCP + REST access with a single bearer key.
  • 126k+ fragrances, 7,500+ brands, normalized 0-1 scores.