mcpskills-trust-scores
收藏资源简介:
MCP Skills信任评分数据集是一个公开的数据集,用于评估AI技能、MCP服务器和npm软件包的信任度。该数据集包含2273个条目,每个条目对应一个已评分的代码仓库或软件包。核心数据包括一个0-10分的复合信任评分(composite)、一个等级标签(tier,如verified、established、new、blocked)以及支持该评分的元数据,这些元数据覆盖了四个评估维度:活跃度(Alive)、合法性(Legit)、稳健性(Solid)和可用性(Usable)。数据集提供了详细的字段,包括标识符(key)、类型(type,如github或npm)、仓库地址(repo)、包名(package)、评分、等级、模式(mode)、限制状态(limited)、GitHub星标数(stars)、分叉数(forks)、许可证(license)、描述(description)、扫描时间(scannedAt)和发现途径(discoveredVia)。数据以CSV和JSON两种格式提供,并遵循CC BY 4.0许可证。此数据集是官方源(mcpskills.io/data/latest.json)的每周自动更新镜像,旨在方便进行数据分析和学术引用。适用于AI代理安全、软件供应链风险评估、开源项目质量分析等场景。
The MCP Skills Trust Score Dataset is a publicly available dataset designed to evaluate the trustworthiness of AI skills, MCP servers, and npm packages. It comprises 2,273 entries, each corresponding to a scored code repository or software package. The core data includes a composite trust score ranging from 0 to 10, a tier label (e.g., "verified", "established", "new", "blocked"), and metadata supporting the score, which covers four evaluation dimensions: Activity (Alive), Legitimacy (Legit), Robustness (Solid), and Availability (Usable). The dataset provides detailed fields including identifier (key), type (e.g., "github" or "npm"), repository address (repo), package name, score, tier, mode, limited status, GitHub stars, forks, license, description, scannedAt, and discoveredVia. The data is offered in both CSV and JSON formats, and is licensed under CC BY 4.0. This dataset is a weekly automatically updated mirror of the official source (mcpskills.io/data/latest.json), intended to facilitate data analysis and academic citation. It is applicable to scenarios such as AI agent security, software supply chain risk assessment, and open-source project quality analysis.
数据集概述:MCP Skills Trust Scores
数据集名称:MCP Skills Trust Scores
许可证:CC-BY-4.0
语言:英语
数据规模:约 1,000 至 10,000 条记录(实际行数:2273 行)
数据集配置:default,数据文件为 mcpskills-trust-scores.csv
来源数据集:原始数据(original)
数据集描述
该数据集是关于 AI 技能、MCP 服务器和 npm 包公开信任评分的数据集。每行代表一个经过评分的仓库或包,包含一个 0-10 的综合信任分数、分级标签以及四个维度(活跃度、合法性、可靠性、可用性)的辅助元数据。
- 数据来源与方法论:https://mcpskills.io/methodology
- 权威实时导出地址:https://mcpskills.io/data/latest.json (以及 https://mcpskills.io/data/latest.csv)
- 许可证要求:使用需注明出处(CC BY 4.0)
- 最近更新:2026-06-15
- 行数:2273
- 同步说明:本数据集每周通过自动化作业从权威实时导出地址同步一次,权威数据源为 https://mcpskills.io/data/latest.json,本数据集为便于分析和引用的镜像。
文件说明
| 文件名 | 描述 |
|---|---|
mcpskills-trust-scores.csv |
完整数据集,RFC 4180 格式的 CSV 文件,许可证注释行以 # 开头。 |
mcpskills-trust-scores.json |
相同数据的 JSON 封装,包含 dataset 元数据、summary(分级计数、平均值)和 rows 数组。 |
列说明
数据集包含以下列:key, type, repo, package, composite, tier, mode, limited, stars, forks, license, description, scannedAt, discoveredVia
composite:0-10 的信任评分tier:分级标签,取值包括verified、established、new、blockedtype:类型,取值为github或npm
使用示例
DuckDB(自动跳过以 # 开头的注释行):
sql
SELECT key, composite, tier
FROM read_csv_auto(mcpskills-trust-scores.csv, comment=#)
WHERE tier = verified
ORDER BY composite DESC
LIMIT 20;
pandas: python import pandas as pd df = pd.read_csv("mcpskills-trust-scores.csv", comment="#") df[df.tier == "verified"].sort_values("composite", ascending=False).head(20)
Hugging Face datasets:
python
from datasets import load_dataset
ds = load_dataset("MisterRiseAbove/mcpskills-trust-scores")
引用
MCP Skills. (2026). MCP Skills Trust Scores [Data set]. https://mcpskills.io/data/latest.json
数据来源于 MCP Skills (https://mcpskills.io),遵循 CC BY 4.0 许可。




