When Claude Code gets 403'd, this plugin escalates

What it does

insane-search is a Claude Code plugin that reads public web pages even when the default fetcher is blocked. When a request hits a 403, CAPTCHA, or WAF, it automatically tries a sequence of public-access routes—platform APIs and RSS feeds first, then TLS impersonation, then a real headless browser—until it returns clean text. It auto-installs its own dependencies and asks for no API keys or proxy setup.

The interesting bit

The escalation logic is the star: a Phase 0→3 scheduler that only brings out heavy artillery—like curl_cffi or a headless browser—when lighter public endpoints fail. It also snoops on the headless browser’s network traffic to reuse the site’s own internal JSON APIs. Under the hood it’s largely orchestrating battle-tested tools rather than reinventing the wheel, which is probably why it actually works.

Key highlights

• Covers X, Reddit, YouTube, HN, LinkedIn, and any site with a public /rss or feed
• Auto-installs curl_cffi, yt-dlp, and friends on first use; zero configuration
• Builds a full browser identity (TLS fingerprint, cookie warming, referer chain) instead of just swapping User-Agent
• Surfaces titles, summaries, and prices from OGP / JSON-LD even on partially blocked pages
• Stops cleanly at login walls and paywalls, reporting “authentication required” instead of crashing through

Caveats

• Deliberately halts at authentication barriers; paywalled or logged-in content is out of scope
• If a platform seals off every public route, even the headless browser can’t help

Verdict

Use this if you want Claude Code to research public forums, videos, and news without juggling API keys or manual workarounds. Skip it if your workflow lives behind authentication or if you already have dedicated scraping infrastructure that handles the same escalation dance.