Claude Code Documentation Tool

Website: costiash.github.io/claude-code-docs — live index stats, try the search in your browser, and watch a grounded hook build step by step.

Every official Claude doc page, always live, never redistributed. Every page across platform.claude.com and code.claude.com, re-indexed every 3 hours — and your Claude reads the actual current page, not a stale copy or a guess from training data.

/plugin marketplace add costiash/claude-code-docs
/plugin install claude-docs@claude-code-docs

Two commands. No Python, no MCP server, no API keys — the client is pure bash + curl + jq.

The Architecture Nobody Else Uses

Most documentation tools pick one of two designs: mirror the docs into a repo (instantly stale between syncs, redistributes content that isn't theirs, ~100 MB clones) or serve them through an MCP server (another process to run, network round-trips per query, answers that flood your context window).

This tool does neither. The repository commits only metadata — a page manifest and a prose-free search index, about 3 MB total. Your machine fetches each .md page directly from Anthropic's servers, on demand, into a local cache. Claude searches the index first and reads only the pages it needs.

	Web search	Docs mirrors	MCP docs servers	This tool
Freshness	Search-engine lag	Stale between syncs	Usually current	Live page, every read
Claude docs coverage	Hit-or-miss	Often partial	Generic libraries	Every page, both sites
Runtime dependencies	None	git + disk	MCP server + config	bash, curl, jq
Context cost per answer	Whole pages of noise	Full file reads	Large tool payloads	Index hit → one targeted page
Redistributes Anthropic's docs	No	Yes, wholesale	Varies	Never — you fetch from the source
Clone size	—	~100 MB	—	~3 MB

The compliance point is not a footnote: this repo contains zero documentation prose — not at the tip, not anywhere in its history. What you install is a map; the territory always comes fresh from Anthropic.

Without claude-code-docs	With claude-code-docs
Claude guesses from training data	Claude reads the latest official docs
Broken or outdated URLs in answers	Correct platform.claude.com / code.claude.com links
"I think the API works like..."	"According to the documentation..."
You verify answers manually	Answers cite specific doc pages

What You Get