Features
📚 Two Libraries
- Public Library - Thousands of curated prompts, agents, skills, guides and system prompts, in
site/library/
- My Library - Your own prompts, one markdown file each, in a folder on your disk
📝 Your Prompts Are Files
- Create, edit and delete prompts in the app, or in any text editor
- Copy any public prompt into My Library with one click
- Tags, sections and categories live in each file's frontmatter
- Point
MY_LIBRARY_DIR at a synced or git folder and your library is backed up
📦 Skill Packs
- Bundles of Agent Skills, one per skills folder plus hand-picked themes, added to or removed from My Library in one step
- Download any skill or pack as a ZIP
🔌 MCP Servers
- One card per Model Context Protocol server: maintainer, licence, auth, and install lines for Claude Code and any
mcpServers client
- Install lines read from each server's own README, re-checked monthly
🤖 Use It From An Agent
- A built-in MCP server: Claude Code, Codex, Cursor or any MCP client can search the library, read any item, pull a skill with its files, and save to My Library
- Featured prompts and your own show up as slash commands
🎨 Modern UI
- Responsive design (mobile, tablet, desktop)
- The org house style shared with ai-automation-tools.dev: sections in a colour rail whose current tab joins the categories panel, a Light | Dark switch, and a Settings dialog with twelve themes that wash the whole page in a colour (Sky, Midnight, Forest, Ember, Wine, Ocean, Slate, Sand, Solar, Synthwave, Sakura, Mono), all remembered per browser; per-section and per-category colours
- ⌘K command palette across every prompt, section and action
- Fuzzy search with Fuse.js
- Title-prioritized search ranking (title starts-with/contains first)
🔒 Nothing To Sign Up For
- No accounts, no database, no API keys, no
.env required
- The server listens on
127.0.0.1 only
- The online demo is the same app in read-only mode
Quick Start
The app lives in site/. Run every npm command from there: the repo root has no package.json.
git clone https://github.com/ai-automation-tools/promptbook.git
cd promptbook/site
npm install
npm run dev # or: npm start (build once, then serve)
Open http://localhost:3010. Your prompts are saved to site/my-library/ (gitignored).
Optional settings
Copy site/.env.example to site/.env to change a default:
| Variable | Default | What it does |
|---|
MY_LIBRARY_DIR | my-library | Where My Library is kept. Relative to site/, or absolute. |
HOST | 127.0.0.1 | Address to listen on. There are no accounts, so read this before opening it up. |
READ_ONLY | false | true serves the Public Library only (My Library is neither listed nor served) and refuses every write. |
ALLOWED_HOSTS | (none) | Extra hostnames the server answers to, comma-separated, for a proxy such as tailscale serve. Any other name gets 403, which blocks DNS rebinding. See this. |
Checks
npm run lint # tsc --noEmit
npm run test:routes # the API route table is intact
npm run test:my-library # the My Library file store
npm run test:read-only # read-only mode hides My Library
npm run test:prompt-content # a failed prompt load never falls back to the preview
npm run test:markdown-file # a downloaded .md reads back unchanged
npm run test:preview # badge rows do not become the card blurb
npm run test:mcp # the MCP server, end to end over stdio
npm run test:safe-path # ../ escapes and links out of library/ are refused
npm run test:browser # after npm run build: the built app in Chrome (needs Chrome installed)
npm run build:index # rebuild api/prompt-index.json after editing library/
npm run bench # latency and event-loop stall of the heavy endpoints (not a test)
node scripts/upstream.test.mjs # provenance frontmatter (run from the repo root)
node scripts/skill-frontmatter.test.mjs # Agent Skills spec compliance
CI runs all of these on every PR and fails if the committed prompt index is stale.
Use It From An Agent
The repo ships an MCP server (site/mcp/server.ts) that reads the same folders as the app.
Nothing else needs to be running. After npm install in site/, add it to your agent:
# Claude Code (use the absolute path to your clone)
claude mcp add promptbook -- npm --prefix /path/to/promptbook/site run --silent mcp
Any other MCP client, as an mcpServers entry:
{
"mcpServers": {
"promptbook": {
"command": "npm",
"args": ["--prefix", "/path/to/promptbook/site", "run", "--silent", "mcp"]
}
}
}
| Tool | Does |
|---|
search_library | Search every section and My Library by words, section, category or tags |
get_item | The full markdown of one item |
list_sections, list_categories | What is in the library, with counts |
get_skill, get_skill_file | A skill's SKILL.md and its support files |
list_skill_packs, get_skill_pack | The Skill Packs and the skills in each |
save_to_my_library, update_my_prompt, update_from_public, delete_my_prompt | Write to My Library (not offered when READ_ONLY=true) |
It also offers MCP prompts: library_prompt takes any library id, and every featured prompt
and every My Library prompt is listed by name, so clients show them as slash commands. It
honours MY_LIBRARY_DIR and READ_ONLY from site/.env. Setup for each client:
docs/mcp-server.md. How it works: docs/features/MCP-SERVER.md.
Documentation
📚 Complete Documentation - Full documentation index
Quick Links
Getting Started
Creating Content
Features
Development
Technology Stack
Frontend:
- React + TypeScript
- Vite (build tool)
- Tailwind CSS
- Lucide React (icons)
- Fuse.js (search)
Backend:
- Express + TypeScript, run with
tsx
- gray-matter (frontmatter for both libraries)
- Markdown files on disk, no database
Hosting:
- Your machine (the app)
- Vercel (the read-only demo)
- GitHub Pages (the homepage)
Project Structure
promptbook/
├── site/ # The app (run npm here; Vercel Root Directory = site)
│ ├── api/ # API handlers (one Express app, also run by Vercel)
│ ├── lib/ # library.ts (shared reader), my-library.ts (file store), safe-path.ts
│ ├── mcp/ # The MCP server (npm run mcp)
│ ├── src/ # React application
│ │ ├── App.tsx # Main React component
│ │ ├── components/ # UI components
│ │ ├── lib/mode.ts # Read-only flag for the UI
│ │ └── main.tsx # Entry point
│ ├── library/ # Public prompt templates (numbered)
│ │ ├── 1_Guides/
│ │ ├── 2_Agents/
│ │ ├── 3_Skills/
│ │ ├── 4_Prompts/
│ │ ├── 5_System_Prompts/
│ │ └── 6_MCP_Servers/
│ ├── my-library/ # Your prompts (gitignored, created on first save)
│ ├── scripts/ # build-prompt-index.js, the API, MCP and browser tests, the benchmark
│ └── server.ts # Local server around the same app api/ exports (port 3010)
├── homepage/ # Static project homepage (GitHub Pages)
├── .github/workflows/ci.yml # lint + route, store and MCP tests + provenance checks + index freshness
├── .gitattributes # LF everywhere, in the repo and the working tree
├── LICENSE, NOTICE # Apache 2.0 for the app; vendored content keeps its own
├── SECURITY.md, CODE_OF_CONDUCT.md, CONTRIBUTING.md
├── docs/ # Project documentation (DISCOVERY.md is the repo-intake ledger)
├── scripts/ # Library tooling: provenance, skills sync, skill packs, discovery, link check
└── README.md
My Library Files
Each prompt is one file, named <slug-of-title>-<6 hex>.md:
---
title: Code review checklist
section: 4_Prompts
category: Coding
tags:
- review
created: '2026-10-21T14:02:11.000Z'
updated: '2026-10-21T14:02:11.000Z'
---
Review this diff for...
Edit them by hand if you like; the app re-reads the folder on every load. Skill packs you add
are ordinary files tagged skill-pack:<packId>. Details in the Setup Guide.
API Endpoints
Prompts
GET /api/prompts?library=public|my - List prompts (add &lightweight=true for metadata only —
no body text at all for the public library)POST /api/prompts/previews - Card blurbs for a batch of ids ({ ids: string[] }, 200 max)GET /api/prompts/:id - One prompt with its full bodyPOST /api/prompts - Create a My Library promptPUT /api/prompts/:id - Update a My Library promptDELETE /api/prompts/:id - Delete a My Library promptPOST /api/prompts/:path/copy-to-my-prompts - Copy a public prompt to My Library
Skill Packs
GET /api/skill-packs?library=public|my - List packs by modeGET /api/skill-packs/:packId - Pack detailsPOST /api/skill-packs/:packId/add-to-library - Add pack to My Library (confirm)DELETE /api/skill-packs/:packId/remove-from-library - Remove pack from My Library (confirm)GET /api/skill-packs/:packId/download - Download pack ZIP
Skills
GET /api/skills/download/:skillPath - Download a skill directory as a ZIP
Every route that writes returns 403 in read-only mode. See docs/features/API.md for detailed documentation.
Contributing
This is a personal project, but suggestions and feedback are welcome!
- Fork the repository
- Create a feature branch (
git checkout -b feature/amazing-feature)
- Commit changes (
git commit -m 'Add amazing feature')
- Push to branch (
git push origin feature/amazing-feature)
- Open a Pull Request
License
The application code is licensed under the Apache License 2.0.
Content under site/library/ is a mix of original work and vendored third-party skills and prompts. Vendored skills keep their own LICENSE file and an upstream: frontmatter block naming the source; see NOTICE and docs/UPSTREAMS.md. Those files are governed by their own licenses, not by this repository's.
Support
Built with ❤️ by Michael Schecht
