Repository navigation
First public release — minimal, stateless docs search for AI coding agents.
Highlights
- 3 MCP tools (~850 schema tokens):
kh_search(ripgrep-ranked snippets + wiki-link connections),kh_read(doc or single section, large files preview-guarded),kh_list(catalog + corpus health header) - 5 CLI subcommands (zero schema cost):
init·new·lint·doctor·uninstall - Stateless by design: search is ripgrep over
docs/at query time — no index, no database, nothing to go stale - Doc-format standard with a self-documenting scaffold (
kh-search new) and a linter (kh-search lint) - Drop-in agent skill (
skills/know-hub/SKILL.md) that teaches a coding agent the tools and the format
Install
pip install know-hub
kh-search initRequires ripgrep on PATH. On Windows, if the kh-search launcher is blocked, use python -m know_hub <subcommand> — the MCP server always launches via python -m and is unaffected.
Docs: Getting Started · Usage · Design