403 is not
an answer.
When WebFetch hits a 403, a WAF or a CAPTCHA on a public page, insane-search switches routes (public APIs and feeds, then browser TLS fingerprints, then a real Chrome) until the content comes back. No API key needed. At logins and paywalls it stops and tells you so.
/plugin marketplace add https://github.com/fivetaku/gptaku_plugins.git /plugin install insane-search@gptaku-plugins
Run `claude plugin marketplace add fivetaku/gptaku_plugins` and `claude plugin install insane-search@gptaku-plugins` in the terminal to install it, then tell me to restart Claude Code.

Same URL, different outcome
Coupang search for "keyboard", first run on September 28, 2026.
All 52 attempts, replayed
A replay of the actual execution trace.
Four sites, run again the same day
This is the second run. The route that works changes from run to run, so Coupang here differs from the video above (route 52, 41s). The raw logs are on GitHub.
Blocked? Next route.
Official routes first
RSS feeds and public APIs the site already exposes.
Look like a browser
Real browser TLS fingerprints, cookies and mobile URLs, in combination.
Use a real browser
If all else fails, it opens Chrome on your machine and reads the rendered page.
To be fair
If you need to crawl thousands of pages on a schedule, a hosted crawler is the better fit. insane-search won't break into login-only pages or paywalls; it tells you it couldn't read them. It's built for the one public page that stopped your conversation.
The moment you're blocked,
it steps in.
/plugin marketplace add https://github.com/fivetaku/gptaku_plugins.git /plugin install insane-search@gptaku-plugins
Run `claude plugin marketplace add fivetaku/gptaku_plugins` and `claude plugin install insane-search@gptaku-plugins` in the terminal to install it, then tell me to restart Claude Code.