io.github.Octen-Team/octen-mcpsearch

octen-mcp

Verified · 8 days ago

URL to LLM-ready markdown — plus per-page category, page_structure, and query-driven highlights.

Install

claude mcp add octen-mcp -- npx -y [email protected]

Requires: OCTEN_API_KEY, OCTEN_API_URL

Our take

An API-key client for the Octen search-and-extract service: search, news_search, and a broad_search that fans a query out into concurrent sub-queries, plus an extract tool that turns up to 20 URLs into LLM-ready markdown annotated with category, page_structure, and query-ranked highlights — the page_structure label ('No Main Content' for login walls and empty shells) is a genuinely useful pre-filter that lets agents discard junk URLs before spending context. Two beta tools (image_search, video_search) are gated behind contacting the company, and headline claims like 62ms average search and 'SOTA embedding models' are the vendor's own numbers, unverifiable from outside. It is a thin TypeScript stdio shim over a commercial API — no key, no function — from a young org (created January 2026, 12 stars, 3 contributors, ~416 weekly downloads) that ships actively and documents client setup thoroughly. The registry has it filed under files, which is wrong: this is a live web search and extraction service, and search is where it belongs. For research-agent builders comparing Tavily/Exa-class search APIs who want structural page labels, not just clean text.

reviewed by hand · 2026-07-05

Something wrong or dead here? Report it