遇见数据集

WebMCP on the Tranco Top 10,000: native-browser measurement data and code

收藏
Zenodo2026-10-28 更新2026-10-29 收录
官方服务:

资源简介:

Data and code for the paper What WebMCP Tools Do the 10,000 Most Popular Domains Expose? A Native-Browser Measurement of Landing and Internal Pages (submitted to ACM Transactions on the Web, Special Issue on the Agentic Web). The study visited web pages with an unmodified Chromium 152.0.7977.82 with native WebMCP support and recorded tools that were both reported by the DevTools WebMCP domain and returned by the page's getTools(); no tool was invoked. It covers a landing-page census of the Tranco top 10,000 (October 11–12, 2026), an internal-page follow-up of 1,785 URLs for 484 operators (October 19), and three control crawls (October 25–27): a same-day revisit of landing and internal pages, a paired comparison of read-only versus ordinary network loading, and a www. fallback for unreachable apex domains. The archive contains the SQLite database with all derived observations and confirmed tools, validated per-target records of every run, all target schedules, the operator/implementation adjudication, the functional codebook of the 126 tool names, the crawler and its pinned configuration, and the scripts that recompute every number and figure in the paper. Raw HTML and script captures are not included. See README.md.

提供机构:
Zenodo
创建时间:
2026-10-27
二维码
社区交流群
二维码
科研交流群
商业服务