An open-source AI agent that brings the power of Gemini directly into your terminal.
Install
npx @pmndrs/docsclaude mcp add docs -- npx @pmndrs/docs{
"mcpServers": {
"docs": {
"command": "npx",
"args": [
"@pmndrs/docs"
]
}
}
}Install command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →docs/getting-started/introduction.mdx
Usage
Read
$ npx @pmndrs/docs # the reader: every library, every page
$ npx @pmndrs/docs drei # straight into one
$ npx @pmndrs/docs drei/performances/instances # straight to a page
$ npx @pmndrs/docs search instanced mesh # one result per line, on stdout
↑↓ drives whichever pane has the focus — the list of pages, or the page itself. ⏎ or ⇥
hands the focus over, esc or ⇥ hands it back, and the lit border says where it is — the
wheel moves whichever pane it points at, focus or no focus. ←→ changes library, b folds the sidebar away, / searches every library at once, o opens the
page in a browser, q quits. Links inside a page are clickable wherever the terminal honours
OSC 8 hyperlinks — iTerm2, Ghostty, WezTerm, Kitty, Windows Terminal.
search is the half a pipe or an agent can use: results read {lib} {path} - {title}, the
shape the MCP server publishes its index in, and nothing found exits 1.
--in drei narrows to one library, --in drei/performances/instances to the matching lines of
one page.
Both read the published llms-full.txt of each library, cached for an hour under
~/.cache/pmndrs-docs — --refresh fetches again. Outside a terminal, a page target is
written to stdout, so npx @pmndrs/docs drei/performances/instances | glow works.
Build
$ cat foo.mdx | npx @pmndrs/docs@latest build # one HTML fragment, on stdout
$ npx @pmndrs/docs@latest build docs out # one .html per .mdx, assets alongside
$ npx @pmndrs/docs@latest build docs out --format website # the whole website, statically exported
Both formats use the same Next.js build and static export. --format fragment (the default)
extracts the compiled MDX from that export: no layout, no stylesheet, no script. Mermaid
diagrams stay fenced blocks, and Sandpack shows its code without the editor.
Rendering a document runs next build, including a single file or stdin. For those inputs,
stdout contains only the HTML; build logs go to stderr. esbuild is used only to package the
CLI for publication, not to load consumer components or render documents.
build --help lists every website option — --libname, --base-path, --icon, --theme-*… Each
one falls back to the environment variable it maps to, the same ones
configuration documents.
To use your repository's React components and shared data in MDX, add docs.config.ts in the current working directory. Use --config path/to/config.ts or DOCS_CONFIG for another location; the flag takes precedence.
See custom components and data for the configuration,
website rendering, and HTML fragment behavior.
Preview a folder of MDX, served
$ curl -sL https://raw.githubusercontent.com/pmndrs/docs/refs/heads/main/preview.sh | \
MDX="docs" \
ICON="🥑" \
sh
Builds the website, then serves it — alongside the MDX folder itself, so relative assets
resolve while you edit. Any option from
configuration goes in the same way, and
VERSION picks a published version of the CLI.
Markdown exports
Similar plugs
The fastest path to AI-powered full stack observability, even for lean teams.
Context7 Platform -- Up-to-date code documentation for LLMs and AI code editors
⭐AI-driven public opinion & trend monitor with multi-platform aggregation, RSS, and smart alerts. 告别信息过载,你的 AI 舆情监控助手与热点筛选工具!聚合多平台热点 + RSS 订阅,支持关键词精准筛选。AI 智能筛选新闻 + AI 翻译 + AI…
An open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM
An open-source, privacy-first, self-hosted knowledge workspace where humans and AI agents work together 开源、隐私优先、自托管的知识工作空间,让人与智能体在此协作