Browser Web Search

Problem

Browser Web Search targets a specific agent workflow: when an AI agent needs to search multiple sites, each platform exposes different endpoints, auth states, and response shapes. Scraping them directly is fragile, hard to parse, and often produces unstable input for downstream model reasoning.

How It Works

It wraps 55 platforms into one CLI search interface, outputs structured JSON without requiring API keys, and supports --jq and --count to trim fields at the source. Public sites can be queried directly, with commands covering search, hot lists, details, comments, and history, including examples such as Toutiao, Zhihu, GitHub, Hacker News, and arXiv. Sites such as WeChat, Xiaohongshu, Weibo, Douban, X, and LinkedIn can reuse an authenticated browser session, but sensitive adapters are denied by default and require explicit opt-in before accessing account-related, comment, history, or user-activity endpoints.

Boundaries

It fits agents that need cross-site retrieval, especially public information, community trends, developer resources, papers, and product databases. Before use, separate public adapters from sensitive ones, reduce returned fields with --jq, and only enable account-facing access when necessary. The underlying package runs JavaScript in an authenticated browser tab context and can theoretically read visible session data. Use an isolated browser profile, pin the version, keep audit logs, and avoid forwarding full responses to other services.