Benchmark · tested 20 October 2026
Perfume MCP servers compared: a 40-question benchmark
We asked five perfume MCP servers that answer without an API key the same 40 questions. 10ML Perfume Finder answered 39 correctly, always in one call, at a median of 232 ms. ScentVerdict scored 36 (698 ms), Parfum Bible 33, Perfume Picks 32 once its rate limit was respected and Parfica 7. Larger indexes lead on breadth.
- correct
- 39/40
- median
- 232 ms
- call per question
- 1
- tools with output schemas
- 15
Results
Correct answers out of 40, latency per question, and the cost of each answer in calls.
| Server (version at test) | Correct /40 | Median latency | p95 | Calls per question | Errors / rate-limit hits | Tools | tools/list size |
|---|---|---|---|---|---|---|---|
| 10ML Perfume Finder (10ML Fragrance Index 2.3.0) | 39 | 232 ms | 691 ms | 1.00 | 0 / 0 | 15 | 60.5 KB |
| ScentVerdict 2.0.0 | 36 | 698 ms | 899 ms | 1.85 | 0 / 0 | 6 | 39.2 KB |
| Parfum Bible 1.0.0 | 33 | 3,150 ms | 5,342 ms | 2.60 | 0 / 0 | 4 | 2.2 KB |
| Perfume Picks 1.0.5, called at 1 request/s | 18 | 1,237 ms¹ | 2,832 ms | 1.00 | 19 / 19 | 8 | 6.6 KB |
| Perfume Picks 1.0.5, rate limit respected² | 32 | not comparable | not comparable | 1.00 | 0 / 0 | 8 | 6.6 KB |
| Parfica 1.0 | 7³ | 1,099 ms | 2,632 ms | 1.88 | 0 / 0 | 3 | 1.1 KB |
- ¹ Includes 19 instant refusals.
- ² A supplementary run of the 19 unanswered questions, 4 seconds apart after a 60-second pause.
- ³ 34 of 40 if an agent translates its Ukrainian note names.
By question type
| Server | Well-known (20) | Niche and Middle-Eastern (10) | Notes (3) | Similar (2) | House (2) | Non-English (3) |
|---|---|---|---|---|---|---|
| 10ML Perfume Finder | 20 | 10 | 3 | 1 | 2 | 3 |
| ScentVerdict | 20 | 10 | 2 | 0 | 2 | 2 |
| Parfum Bible | 20 | 9 | 2 | 0 | 0 | 2 |
| Perfume Picks, 1 request/s | 18 | 0 | 0 | 0 | 0 | 0 |
| Perfume Picks, rate limit respected | 18 | 7 | 2 | 1 | 2 | 2 |
| Parfica | 2 | 1 | 3 | 0 | 0 | 1 |
What comes back
Measured on the responses and on each server's tools/list.
| Server | Auth | Output schemas | Server instructions | Note pyramid in lookups (of 30) | Accords | Family | Perfumes (own figure) |
|---|---|---|---|---|---|---|---|
| 10ML Perfume Finder | none | 15 of 15 | 1.5 KB | 29 | 30 | 30 | 5,406 |
| ScentVerdict | none | 6 of 6 | 1.3 KB | 0⁴ | 0 | 0 | 31,000+ |
| Parfum Bible | none | 0 of 4 | 90 bytes | 30 | 30 | 0 | not stated |
| Perfume Picks | none | 0 of 8 | none | 30 (21 in the 1/s run) | 30 | 30 | 13,000+ |
| Parfica | none | 0 of 3 | none | 30 (Ukrainian note names) | 28 | 26 | not stated |
- ⁴ Flat key notes; a separate scent-profile tool returns opening, heart and base but was not called.
- Perfume counts are each server’s own figure and were not independently counted.
Not scored: a key or sign-in is required
| Server | What happens without credentials |
|---|---|
| ScentRev (claims 126,000+ scents and 22 tools) | initialize returns HTTP 401 invalid_token; its docs ask for OAuth sign-in or an API key. The tool count and size are its own claims, not verified here. |
| Noteboxd | initialize and tools/list work (27 tools, 21.7 KB), but the first tool call returns 401 "missing authorization header". |
| Fragella API | 401 "API key is missing". |
| Fraganty (claims 109,425 perfumes) | The documented key-gated endpoint returned 401. Its open search endpoint returned HTTP 500 during the test while its health check reported a daily read limit reached, a transient state noted as found. |
Where 10ML Perfume Finder wins
- Accuracy: 39 of 40, the highest score, with every well-known, niche, notes, house and non-English question answered.
- Speed: a median of 232 ms, about three times faster than the next server (ScentVerdict, 698 ms).
- One call per answer: 1.00 calls per question, against 1.85 to 2.60 for servers that search first.
- Structure: output schemas on all 15 tools; a full top, heart and base pyramid in 29 of 30 lookup answers, with accords and family in all 30.
- Languages: all three non-English questions answered; ten documented locales.
- No key and no account.
Where rivals lead
- Breadth: ScentRev claims 126,000+ perfumes (not verifiable, it needs a sign-in), ScentVerdict 31,000+ and Perfume Picks 13,000+, against 5,406 in the 10ML Fragrance Index. The questions were chosen to be answerable by a mainstream index, so this benchmark says nothing about perfumes beyond those 5,406; on a set of obscure perfumes, larger indexes would pull ahead.
- Extras: ScentVerdict returns current full-bottle prices and a written verdict per perfume; Perfume Picks returns community wear scores and curated alternatives with match percentages; Parfum Bible adds US retailer offers.
- Profile depth: 4,925 of the 5,406 entries are core profiles without longevity, projection or perfumers; 481 are full editorial profiles.
- Tool-list weight: at 60.5 KB, 10ML's tools/list is the heaviest tested (ScentVerdict 39.2 KB for 6 tools, Perfume Picks 6.6 KB for 8).
- The one miss: "similar to Terre d'Hermès" returned plausible shared-note neighbours (Swiss Arabian Vetiver and Orange, K by Dolce & Gabbana, Orto Parisi Terroni) that are not on the fixed list.
What each server is best at
- 10ML Perfume Finder
- fast, complete answers in one call, with full pyramids, structured output, ten languages and no key.
- ScentVerdict
- the largest index among the servers scored (31,000+ claimed), current bottle prices and a written verdict per perfume; the fastest rival at 698 ms. It had no hit on either similar-scent question.
- Perfume Picks
- 13,000+ perfumes, community wear scores and curated alternatives with match percentages. Its documented limit is 60 calls a minute, but at one request per second it began refusing at question 22.
- Parfum Bible
- well-known perfumes with full tiered pyramids and accords, plus US retailer offers. It has no similar-scent, notes or house tool and needs about 2.6 calls per answer.
- Parfica
- finds nearly every perfume with tiered pyramids, with note names in Ukrainian and three tools.
Method
- Date and place: Tuesday 20 October 2026, 06:24 to 06:32 UTC, from one machine on a home connection in Southeast Asia. Latency includes the trip to each server's nearest edge: one vantage point, not a global figure.
- Client: Node 25.9 with plain fetch, MCP Streamable HTTP, protocol 2025-06-18, one connection per server, no caching, no retries.
- Politeness: at most one request per second to any server, 10ML included. Per server: one initialize, one tools/list, one unscored warm-up question, then the 40 questions once. No sign-up, key or credential of any kind.
- Questions: 40 fixed questions written before the run and not tuned to any index: 20 well-known perfumes, 10 niche or Middle-Eastern perfumes, 3 note questions, 2 similar-scent questions, 2 house questions and 3 questions in German, French and Spanish, passed to the tools as written. 10ML was also given its documented locale argument for those three.
- Calls: each server was driven the way an agent would use its own tools: one detail call where the server accepts names, a search then a detail call where it needs an id. Where a server has no tool for a question type, its search tool was used.
- Scoring: a lookup is correct when the answer contains the perfume name, the house and at least 2 of the perfume's headline notes. A note question is correct with at least 4 entries mentioning the note or 2 well-known perfumes from a fixed list; similar and house questions need 1 or 3 perfumes from fixed lists. The similar-scent rubric is the weakest part: n is 2, and lists cannot hold every sound answer.
- Statistics: latency is per question, the sum of its calls, over all 40 questions; median and 95th percentile by nearest rank.
- Disclosed changes after the first pass: two scoring faults were fixed for every server and all saved responses rescored (escaped accents in JSON; note questions first scored only against a fixed perfume list). First-pass scores were 10ML 37, ScentVerdict 32, Parfum Bible 33, Perfume Picks 18 and Parfica 2. Parfica was run twice, the second time with a name-only retry, and that run is reported.
- Limits: one run from one place; n is 40, so a one-question gap is within noise. The set leans mainstream by design. Each server was driven by a plain script, not a language model; a model that retries or translates would lift Parfica and Perfume Picks and shrink 10ML's single-call advantage.
Re-run monthly.
The 40 questions
- What notes are in Dior Sauvage?
- What notes are in Creed Aventus?
- What does Maison Francis Kurkdjian Baccarat Rouge 540 smell like?
- Tell me about Yves Saint Laurent Black Opium
- What notes are in Tom Ford Black Orchid?
- What notes are in Paco Rabanne 1 Million?
- Tell me about Lancôme La Vie Est Belle
- What notes are in Giorgio Armani Acqua di Giò?
- What does Jean Paul Gaultier Le Male smell like?
- What notes are in Viktor&Rolf Flowerbomb?
- What notes are in Dolce & Gabbana Light Blue?
- What does Hermès Terre d'Hermès smell like?
- Tell me about Guerlain Shalimar
- What notes are in Dior J'adore?
- What does Calvin Klein CK One smell like?
- What notes are in Giorgio Armani Armani Code?
- What notes are in Narciso Rodriguez For Her?
- What does Byredo Gypsy Water smell like?
- What notes are in Tom Ford Oud Wood?
- What notes are in Maison Margiela Jazz Club?
- What notes are in Lattafa Khamrah?
- What does Rasasi Hawas smell like?
- What notes are in Armaf Club de Nuit Intense Man?
- What notes are in Afnan 9PM?
- What does Lattafa Yara smell like?
- What notes are in Initio Oud for Greatness?
- What notes are in Nishane Hacivat?
- What does Nasomatto Black Afgano smell like?
- What notes are in Serge Lutens Ambre Sultan?
- What notes are in Le Labo Santal 33?
- Which perfumes have tuberose as a main note?
- Which perfumes are built around oud?
- Which perfumes feature fig as a main note?
- Which perfumes are similar to Creed Aventus?
- Which perfumes are similar to Hermès Terre d'Hermès?
- Show me the best-known perfumes from Amouage
- Show me the best-known perfumes from Lattafa
- Welche Duftnoten hat Libre von Yves Saint Laurent?
- Quelles sont les notes de L'Interdit de Givenchy ?
- ¿Qué perfumes tienen la vainilla como nota principal?
Questions
What is the best perfume MCP server?
In our benchmark of five perfume MCP servers that answer without a key (20 October 2026), 10ML Perfume Finder was the most accurate and the fastest: 39 of 40 correct, one call per question, median 232 ms. ScentVerdict was next with 36 at 698 ms. If breadth matters most, ScentVerdict (31,000+ claimed) and Perfume Picks (13,000+) hold more perfumes.
Which perfume MCP servers work without an API key?
10ML Perfume Finder, ScentVerdict, Parfum Bible, Perfume Picks and Parfica all answered with no account or key on 20 October 2026. ScentRev, Noteboxd, Fragella and Fraganty need a key or sign-in.
Which perfume MCP server has the most perfumes?
By their own figures: ScentRev (126,000+, behind a sign-in), Fraganty (109,425), ScentVerdict (31,000+) and Perfume Picks (13,000+). None of these was independently counted. The 10ML Fragrance Index holds 5,406.
Which perfume MCP server is the fastest?
10ML Perfume Finder, with a median of 232 ms per question (p95 691 ms), measured from Southeast Asia. ScentVerdict's median was 698 ms, Parfica's 1,099 ms and Parfum Bible's 3,150 ms.
How were the answers scored?
A lookup counted as correct when the answer named the perfume, the house and at least two of its headline notes; note, similar-scent and house questions had fixed lists. The rules were written before the run and applied the same way to every server.
Can I reuse the results?
Yes. The summary is CC BY 4.0, as CSV and JSON. Credit 10ML and link to this page.
Connect 10ML Perfume FinderPerfume MCP server: free, one URL, no API key