Perfume MCP servers 2026: 40-question benchmark review
An in-depth benchmark review comparing 5 perfume MCP servers using 40 standard questions on October 20, 2026.
Stock photo for illustration only, not from the actual event
- Benchmark tested perfume MCP servers using 40 standard questions on October 20, 2026.
- 10ML Perfume Finder achieved the highest score with 37 out of 40 points.
- Only 5 servers were scored as they required no API key or sign-in.
- Summary benchmark data is available as CSV and JSON under CC BY 4.0.
When searching for a perfume MCP server, developers often find a handful of options accompanied by very little behavioral data. To address this, on October 20, 2026, a rigorous test was conducted running the exact same 40 questions against five perfume MCP servers that respond without requiring an API key or sign-in
Stock photo for illustration only, not from the actual event
Prior to disclosing the metrics, the evaluators made a transparent conflict-of-interest disclosure: one of the tested servers, 10ML Perfume Finder, was built by the testing team. To maintain integrity, the methodology, rubric, 40 questions, and limitations have been published openly on 10ml.com.
Evaluated based on responses and each server's tools/list capabilities, the initial pass scores recorded were 10ML at 37, ScentVerdict at 32, Parfum Bible at 33, Perfume Picks at 18, and Parfica at 2. Parfica was given a second run with a name-only retry, which is the data officially reported.
Model Context Protocol (MCP) servers bridge AI applications with external structured data sources. Benchmarking these endpoints with a fixed questionnaire helps developers evaluate reliability, handling of rate limits, and response accuracy. However, performance can vary depending on client-side translation layers and automated retry logic not present in simple benchmarking scripts.
Every server was driven by a plain script from a single location. With n = 40, a single-question gap remains well within statistical noise. Two scoring faults discovered after the initial pass—escaped JSON accents and note-based questions evaluated against a rigid perfume list—were fixed across all saved responses for fairness
Stock photo for illustration only, not from the actual event
The team plans to re-run the benchmark on a monthly basis. Developers maintaining perfume servers or wishing to suggest additional question categories are encouraged to provide feedback, with complete CSV and JSON datasets accessible via the benchmark page.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment