I've been using AI agents like OpenCode, Claude Code, and Cursor for months. They're great with code, but when they need to search or browse the web, things get complicated: Cloudflare blocks them, JavaScript-heavy sites don't load, APIs cost money.
So I built browser-search.
It's three open source tools orchestrated by a skill, fully self-hosted:
**SearXNG **— metasearch engine that queries dozens of search engines at once
**Camofox **— full browser via REST API, always warm, for browsing and interacting
**CloakBrowser **— stealth browser for when the site has Cloudflare, Akamai, or DataDome
The agent decides which tool to use. Zero human intervention. Zero API keys. Zero subscriptions.
What makes it different:
It's a skill, not a plugin — works with any agent that can read instructions
Automatic navigation escalation: if Camofox gets blocked, it switches to CloakBrowser
Deep Research mode: the agent is instructed to go beyond surface-level answers, cross-verify sources, cover every aspect
Integrated Readability.js for clean article extraction (~70% token savings)
The SKILL.md is plain text — fork it, tweak it, make it yours
Built-in security. Browser-search is designed to be safe to install and use, including SSRF protection, script sandboxing, rate limiting, and path traversal blocks.
MIT licensed on GitHub: [https://github.com/Johell1NS/browser-search](https://github.com/Johell1NS/browser-search)
If you try it, let me know. If you make it better, even more so. If you don't need it, share it with someone who might. Every star, comment, or pull request is welcome — that's what makes open source great.