Deep search infrastructure for AI agents

What it is

An ordinary search tool gives an agent one engine and a summary. This one gives it sixteen engines mapped to what each is good at: general web, the small web, archives, open directories, device search, even the dark web. It is built as an MCP server. Any agent that speaks the protocol can use it.

The social side runs sixteen platforms, from Reddit and HackerNews to Telegram and 4chan, with Reddit using a documented three-route race so a rate limit never means a dead end.

Under the hood

• Sixteen engines, each wrapped with its own client, rate policy and failure mode: web, Marginalia small-web, Anna's Archive, Wayback undelete, open-directory file search, Shodan and Censys, plus more.
• Tor onion search with a maintained source catalog, liveness checking and a discovery CLI that runs out of band.
• Sixteen social platforms through one interface, with per-platform concurrency caps and honest rate-limit reporting.
• An SSRF-guarded HTTP layer so an agent cannot be walked into internal addresses through a search result.
• Actively maintained, with an evidence corpus documenting what each engine actually returns.

Why it exists

Agents fail at research for the same reason people do: they stop at the first source. The fix is not a smarter prompt, it is more doors. Each engine here was added because a real question needed it. The catalog of onion sources exists because searching the dark web blind wastes hours.

Stack