Hi Hacker News. I built this primarily because, from my experience, the alternatives were all either paid cloud APIs, or simply didn't work in my tests. Or were simply far too heavy. Whoogle sadly has gone on to a better place. Google has been clamping down on HTML scraping, and brave ended their free api quota.
Serica is one ~16MB static binary. `cargo install serica-search`, run `serica`, and then you have two tools at your disposal, accessible either via REST or MCP.
`GET /api/v1/search?q=...` - duckduckgo results as clean JSON (pretty much the only engine i found that reliably worked with HTML scraping. This limitation is brought on because i opted not go utilize headless browsers, which could in theory work with JS and google by proxy, but that has overhead.
`GET /api/v1/extract?url=...` - extracts the main text from (mostly) any article URL into Markdown, ads and nav stripped (trafilatura-based). Known limitations here are SPA / JS heavy sites frequently break. Thats the tradeoff with not using headless browsers. Implementing a fallback would have inflated the projects scope, so i opted to avoid that.
No keys, no database, (mostly) no bloat. Redis caching exists but is compile-time. Off by default.
SSRF hardening (redirect re-validation on every hop) is also present.
Hi Hacker News. I built this primarily because, from my experience, the alternatives were all either paid cloud APIs, or simply didn't work in my tests. Or were simply far too heavy. Whoogle sadly has gone on to a better place. Google has been clamping down on HTML scraping, and brave ended their free api quota.
Serica is one ~16MB static binary. `cargo install serica-search`, run `serica`, and then you have two tools at your disposal, accessible either via REST or MCP.
`GET /api/v1/search?q=...` - duckduckgo results as clean JSON (pretty much the only engine i found that reliably worked with HTML scraping. This limitation is brought on because i opted not go utilize headless browsers, which could in theory work with JS and google by proxy, but that has overhead.
`GET /api/v1/extract?url=...` - extracts the main text from (mostly) any article URL into Markdown, ads and nav stripped (trafilatura-based). Known limitations here are SPA / JS heavy sites frequently break. Thats the tradeoff with not using headless browsers. Implementing a fallback would have inflated the projects scope, so i opted to avoid that.
No keys, no database, (mostly) no bloat. Redis caching exists but is compile-time. Off by default.
SSRF hardening (redirect re-validation on every hop) is also present.