This is an early release preview. You may encounter bugs.
Browser Search logo
Unclaimed

Tool interface research

Browser Search

Agent skill orchestrating SearXNG, Camofox, and CloakBrowser for self-hosted web search and anti-bot browsing

A 80/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
No votes yet

01 / About

What Browser Search is.

browser-search is an agent skill that gives coding agents such as OpenCode, Claude Code, Cursor, and OpenClaw a self-hosted pipeline for searching the web and reading the results. It wires together three components: SearXNG as a metasearch engine for the search phase, Camofox as a REST-driven Firefox browser for ordinary sites, and CloakBrowser as a stealth Chromium for sites behind anti-bot protection. All three run on your own machine through Docker and npm, so there are no API keys or per-query charges.

The skill restricts the model to a fixed set of deterministic scripts: each tool is described in natural language in SKILL.md, but execution goes through Node scripts with defined commands and output formats, so the model does not compose commands itself. A typical run searches with SearXNG, opens result URLs in Camofox, and escalates to CloakBrowser only when a Cloudflare, Akamai, or similar challenge blocks the page. The agent chooses the tier on its own.

Camofox keeps an always-warm browser and returns accessibility snapshots with stable element references, plus Readability.js extraction that strips navigation and ads from articles. CloakBrowser launches a fresh Chromium per request with fingerprint patches, proxy and GeoIP options, and automatic challenge detection that waits for Cloudflare, Turnstile, DataDome, Akamai, Imperva, PerimeterX, or DDoS-Guard checks to resolve before extracting content. Because SKILL.md is plain text, you can edit the rules or swap in other tools.

Features

  • Metasearch phase: SearXNG on localhost:8080 queries Google, Bing, DuckDuckGo, Wikipedia, and others at once and returns JSON titles, snippets, and URLs
  • REST-driven browsing: Camofox on localhost:9377 creates tabs, navigates, clicks, scrolls, and evaluates JavaScript against a Firefox instance
  • Article extraction: a readability command uses Mozilla Readability.js to return clean article text with a snapshot fallback
  • Stealth escalation: cloak-fetch.mjs and cloak-script.mjs run CloakBrowser with challenge detection, retry, --proxy, --geoip, and Markdown output options
  • Deterministic scripts: the model invokes fixed Node commands rather than composing shell or API calls, which constrains hallucinated tool use
  • Deep Research mode: a search-first workflow that requires claims to be verified against live sources from several angles
  • Editable rules: SKILL.md is plain text you can trim, extend, or translate for other agent environments
  • Low-resource footprint: developed and tested on a Raspberry Pi, with CloakBrowser 0.5.x adding Linux arm64 support

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

03 / Build

Build with Browser Search.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Browser Search →

Browse the catalogue