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Broken or outdated URLs in answers	Correct platform.claude.com / code.claude.com links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

• Auto-discovery — ask anything about Claude Code, the API, SDKs, or prompt engineering and Claude reads the relevant docs before answering. No prefix, no command. This is the feature you'll forget is running — your answers are just correct.
• /docs skill — explicit lookups when you want them: /docs hooks, /docs extended thinking, /docs Agent SDK sessions
• Interactive Courses — turn any topic into a self-contained animated HTML course (below — this one's unique)
• Changelog Reports — visual HTML reports of what changed in the docs, with one-click course generation per entry
• Token-efficient by design — search runs in shell against a prose-free index; Claude's context only ever pays for the specific pages it reads
• Session-start auto-updates — metadata syncs and changed pages re-fetch in the background every session. No cron jobs, no manual pulls.

On your first session after install, Claude clones the ~3 MB metadata repo to ~/.claude-code-docs/, warms the page cache in the background, and activates the skills. That's the whole setup.

Interactive Courses — Learn Claude by Doing

No other docs tool does this. Ask about any Claude topic and get a beautifully crafted, interactive HTML course — animated protocol diagrams, hands-on quizzes, code-with-English translations, in a dark Obsidian & Amber theme. One self-contained file, zero setup, opens right in your browser.

/docs --course hooks          # Generate a course on Claude Code hooks
/docs --course tool use       # Deep dive into the Tool Use API
/docs --course prompt caching # Master caching with visual explanations

Or just ask naturally after any docs response — Claude will offer to create a course on whatever you just looked up.

https://github.com/user-attachments/assets/e36ae4c1-2ee6-4932-b0a5-3463cd20e012

What's inside a course:

• 4-7 scroll-based modules with a progressive learning arc
• Protocol Conversations — animated chat-style visualizations of how Client, API, Claude, and tools actually exchange messages
• Code translations — real API examples from the docs with line-by-line plain-English explanations
• Quizzes that test application — "What would you use for this scenario?", not "Define this term"
• Glossary tooltips on every Claude-specific term
• Obsidian & Amber theme — Instrument Serif headings, warm amber accents, grain textures, glass effects. Not your typical tutorial.

Courses are saved to ~/.claude-code-docs/courses/ so you can revisit them or share them with your team.

Changelog Reports — the Docs→Learning Loop

Anthropic's documentation changes daily. Stay ahead of it:

/docs --report          # Last 7 days of changes
/docs --report 24h      # Last 24 hours
/docs --report 30d      # Last 30 days

You get a visual HTML report of every added, changed, and removed page — grouped by category, with summaries. And each entry carries a "Create Course" button: click, paste into Claude Code, and deep-dive into whatever just changed. Docs monitoring and learning in one loop.

Usage

Direct Lookups

/docs hooks              # Claude Code hooks
/docs mcp                # MCP server configuration
/docs agent sdk python   # Agent SDK Python guide
/docs --course hooks     # Generate an interactive course on hooks
/docs --report