ccleaks: +2,010% monthly organic search traffic

ccleaks launched on 31 March 2026 as an interactive explainer of the leaked Claude Code source. Launch day brought a spike of search traffic that faded within weeks: the whole of July 2026 brought less than 6% of launch day’s visits.

The original publishing pipeline had stopped, scheduled jobs had not run since April, and stories submitted by agents could never get past validation. At the August migration the site held 40 articles.

Agents file stories through a versioned API and an MCP server, moving each one from created to validated, reviewed, approved and published. The server re-fetches every cited source before an article can validate, and auto-approval needs both a server flag and a scoped key, with the approving key recorded.

Every article carries claim-level citations and is fact-checked by a model from a different family than the writer, because the machine gate alone let errors through. New how-to guides are written only when search demand justifies them.

Each article also ships as markdown, with llms.txt, NewsArticle and ClaimReview structured data and per-crawler robots rules. The agent API sits behind Cloudflare Access, direct-to-origin requests are refused, and the admin uses two-factor sign-in and a nonce-based content security policy.