The Prompt Library That Stops You Starting From Scratch
prompts.chat has 164,000 GitHub stars and 108,000 community prompts. Here's how to use it in a real workflow — browsing, CLI, MCP, and self-hosting for privacy.
Every prompt session starts the same way. You open Claude, or ChatGPT, or whatever model you're in that day, and you type something like "act as a..." — and then you stop. You know what you want. You've done this before. But you're starting from nothing again, and you know the first draft will be mediocre until you've iterated it three or four times into something that actually works.
That's a solvable problem. Most of the prompts you'll ever write exist somewhere already, tried and refined by someone who had the same goal. prompts.chat is where that work lives.
What It Is
prompts.chat is the largest open-source prompt library in existence. It launched in December 2022 as "Awesome ChatGPT Prompts" — a simple GitHub repo with a CSV file — and it compounded from there. Today it has 164,000 GitHub stars, 108,911 community-written prompts, and endorsements from the co-founders of OpenAI and the CEO of Hugging Face. Forty academic papers cite it. Harvard and Columbia reference it.
None of that matters much for a working design engineer. What matters is that when you need a prompt for UX research synthesis, or component API critique, or a code review framing, there is almost certainly one already written, voted on, and refined by someone who uses it daily.
The data is CC0 — public domain. You can use every prompt without attribution, modify freely, commercialize without restriction. The source code is MIT. Both are available on GitHub.
Using It in the Workflow
Browsing the web interface. The fastest path is prompts.chat/prompts. No account required. There's AI-powered semantic search, browsing by category, and a voting system that surfaces what actually works. Prompt categories span coding, writing, UX research, education, business, language translation, and fifty more specialized areas. If you need a starting point for prompting an LLM as a UX critic, a code reviewer, or a documentation writer, search first.
The CLI. If you live in the terminal, npx prompts.chat runs an interactive prompt browser without opening a browser. Useful when you're mid-session in Claude Code and want to pull a prompt without breaking flow.
npx prompts.chat
MCP integration. This is where it gets practical for Claude Code users. prompts.chat ships an MCP server. Add it to your Claude Code config and you get search and retrieval tools directly inside your agent sessions. You can pull a prompt from the library, fill its variables, and feed it into the model without leaving the IDE.
Remote MCP (recommended — no local install):
{
"mcpServers": {
"prompts.chat": {
"url": "https://prompts.chat/api/mcp"
}
}
}
Local MCP (runs on your machine):
{
"mcpServers": {
"prompts.chat": {
"command": "npx",
"args": ["-y", "prompts.chat", "mcp"]
}
}
}
Once connected, you get tools: search_prompts, get_prompt with variable substitution, save_prompt to publish back to the community, and improve_prompt for AI-assisted refinement. It fits naturally next to the MCP tools you're already using.
Variables. Many prompts use placeholder syntax like ${Position:Software Developer}. When you retrieve a prompt through the MCP or web UI, variables are surfaced for you to fill before pasting. That's the difference between a template that forces you to hunt for substitution slots and one that hands them to you.
Self-Hosting for Private Work
The web version is public by default. If you're writing prompts for a client, for internal design system workflows, or for proprietary processes you don't want indexed, self-hosting is the right call.
Docker is the fastest path:
git clone https://github.com/f/prompts.chat.git
cd prompts.chat
docker compose up -d --build
# Visit: http://localhost:4444
That gives you the full platform — browsing, search, user submissions, version control (a change request system modeled on GitHub PRs), voting, and multi-language support — running on your own infrastructure. A PostgreSQL database, Node.js 24, and you're running in under five minutes.
The self-hosted instance is configurable down to the theme and branding. You can toggle features on and off: private prompts, AI-powered semantic search (requires an OpenAI key), comments, categories, the MCP integration. There's even a white-label mode that strips prompts.chat branding entirely for client deployments.
Who Gets the Most Out of It
Design engineers and UX researchers. If you're using AI to synthesize research notes, generate interview questions, critique UI copy, or review component APIs, the library has prompts for all of it. Starting from a community-validated version is faster than iterating your own from zero.
Developers using AI daily. The coding prompts cover code review, documentation, debugging, architecture discussion, and role-based modes (Ethereum developer, Linux terminal, security auditor). These are the use cases where prompt quality directly affects output quality — and where a good starting prompt is worth more than you'd expect.
Teams who want a shared library. Self-hosted prompts.chat is the internal prompt database your team actually keeps current. The change request system means prompts can be versioned and improved collaboratively without degenerating into a Google Doc that's six months stale.
Educators and learners. prompts.chat includes an interactive book on prompt engineering — 25 chapters, free, from basics through chain-of-thought and few-shot learning. It's a legitimate curriculum, not marketing content.
The Real Value
The library is not a substitute for knowing how to write a good prompt. A prompt that someone else wrote for a slightly different task will need editing. Variables won't match your domain. The framing will be generic in places you need specificity.
But starting from something real — something that's been used, voted on, iterated — is faster than starting from nothing. The blank input problem is a friction problem, not a skill problem. 108,000 prompts means the starting friction is almost always zero.
The best use of prompts.chat is reference, not replacement. Browse before you write. Pull the pattern. Make it yours.
Get started: prompts.chat — web interface, no account required. Repository and self-hosting docs at github.com/f/prompts.chat.