prompts.chat is a free, open-source library of prompts for writing, coding, images and research. On its own it's a list you'll forget to open. Wired in as an MCP plugin, Claude searches it and picks the right one for the task itself, so you stop keeping a messy note of prompts and pasting them in one by one.
Think of prompts.chat as a big shared notebook. Someone else has already written thousands of good prompts for writing, coding, images and research, and left it open for anyone. It's free and open source, and you can even keep your own private copy.
The catch is that a notebook only helps if you remember to open it. Most people keep a scrappy version instead, a note on their phone or a doc of half-finished prompts, and dig through it by hand. You forget it exists, copy in the wrong version, and rewrite the brief you wrote last week. The fix isn't a tidier note. It's letting Claude read the notebook itself, so it grabs the right page mid-conversation without you leaving the terminal.
prompts.chat is a real, open-source project at github.com/f/prompts.chat, with an official Claude Code plugin documented in that repository's own CLAUDE-PLUGIN.md file, which this guide is built from directly.
If you already run Claude Code plugins, skip to the install section. If not, here's the vocabulary in plain terms.
These are the exact commands from prompts.chat's own Claude Code plugin documentation. Run them inside a Claude Code session, not in a plain terminal.
/plugin marketplace add f/prompts.chat /plugin install [email protected]
The first line points Claude Code at the prompts.chat marketplace. The second installs the plugin from it, which wires up the MCP connection, two slash commands and two agents in one step.
Skip this unless you want Claude to save or improve prompts back to your prompts.chat account, not just search the public library. Grab a key from prompts.chat/settings, then set it as an environment variable before you launch Claude Code:
export PROMPTS_API_KEY=your_api_key_here
prompts.chat also exposes itself as a plain MCP server, for tools that don't support Claude Code's plugin marketplaces. Point any MCP-compatible client at the remote server, or run it locally if you'd rather not depend on their hosting:
{
"mcpServers": {
"prompts.chat": {
"url": "https://prompts.chat/api/mcp"
}
}
}{
"mcpServers": {
"prompts.chat": {
"command": "npx",
"args": ["-y", "prompts.chat", "mcp"]
}
}
}/prompts.chat:prompts cold outreach email /prompts.chat:prompts logo concept --type IMAGE /prompts.chat:prompts code review --category coding
/prompts.chat:skills pdf
Typing /prompts.chat: in Claude Code shows the prompts and skills commands autocomplete. A search returns real results from the library, not an error about a missing MCP connection.
Slash command not found: the plugin didn't finish installing, so run the two install lines again and fully restart Claude Code. Search runs but returns nothing useful: your query is too broad, so add --category or --type to narrow it. Saving a prompt fails: check PROMPTS_API_KEY is set in the terminal session Claude Code is running in, not a different one.
The trap is treating the library as a shortcut around your own briefing. A library prompt still needs your context filled in. And ten thousand prompts sitting in a browser tab you don't open is worth less than twenty your agent can reach mid-task. Wiring the library in as MCP turns "I should look that up" into something Claude just does.
A prompt you have to go find is a prompt you'll stop using.
Sources and honesty notes: this guide is built from prompts.chat's own repository at github.com/f/prompts.chat, including its CLAUDE-PLUGIN.md file for the exact install and usage commands. The size of the library ("10,000 prompts") is a figure commonly quoted for this project and is stated here as a widely repeated claim, not a count I verified myself line by line. Commands and MCP config shown are current as of this write-up and may change between versions, so check the repository if something doesn't match.
The Audacity to Build is my community for people doing exactly that: building real AI systems alongside me, step by step. Join the waitlist and I'll tell you the moment doors open.
If this was useful, follow me. I share what I am actually building, the tools that survive my own testing, and the ones I quietly cancel. New plays every week.
Pop in your name and email and the rest of the guide is yours, along with the occasional new guide when I publish one. No spam, ever.