A huge open-source prompt library that plugs into Claude Code. But the prompt your agent gets is not the best one, it is the newest
prompts.chat is the repo formerly known as Awesome ChatGPT Prompts. With 171k stars on GitHub it sits among the most starred repositories in the world. Prompts for writing, code, images, video and audio plus agent skills in one place; the code is MIT and the content is CC0, so you can use it however you want. The real trick is the MCP server: connect the library to Claude Code and your agent searches for and pulls prompts itself. I checked four things. First, search ordering: I read the server code, and results come back by date added, not by votes or quality. So the first prompt your agent sees is often a zero-vote attempt someone posted today. Second, the difference between the plugin and plain MCP. Third, accounts: not needed to read, needed to save. Fourth, skills: the community writes them, and they can include scripts. Below: two install routes, how to get your agent to find the right prompt, and the actual size of the library.
The library itself
You can browse and copy prompts on the site without an account. They are split by type: text, image, video, audio and agent skills. Some prompts also come with a ready workflow link. The code is MIT licensed and the prompt content is CC0, so you can use it in commercial work without attribution.
What is extra below: the two ways to connect it to Claude Code and which one to pick, the ordering trap in your agent's search results, when you actually need an account, and how big the library really is today.
Two ways to connect it to Claude Code
Plugin or plain MCP
Plugin
Everything comes bundled
- The MCP server connects automatically
- Two slash commands: prompt search and skill search
- Skills that kick in on their own when you mention prompts
- Two agents for managing prompts and skills
Plain MCP
Just the tools
- Added with one command, installs nothing else
- Your agent uses the tools when you ask
- Works with the same address in other editors like Cursor and Windsurf
If you use Claude Code, the plugin is the comfortable route. Type these two lines inside Claude Code, not in your terminal.
/plugin marketplace add f/prompts.chat
/plugin install prompts.chat@prompts.chatOnce installed you get two commands. They take category, tag or type filters too; the type filter helps when you are looking for image or video prompts.
/prompts.chat:prompts code review
/prompts.chat:prompts midjourney --type IMAGE
/prompts.chat:skills testing automationIf you only want the tools, plain MCP is enough. This command runs in your terminal. The server is remote, nothing gets downloaded to your machine.
claude mcp add --transport http prompts.chat https://prompts.chat/api/mcpIf you use a different editor, adding the same address to its MCP config file is enough.
{
"mcpServers": {
"prompts.chat": {
"url": "https://prompts.chat/api/mcp"
}
}
}The prompt your agent finds is not the best one
Once connected, your agent gets ten tools: searching, fetching and saving prompts and skills, plus one that improves a prompt. The idea of the agent picking the prompt it needs is sound, but it is worth looking at what it is picking from.
How the search works
Search results come back by date added, newest first. There is no ordering by votes. I tried it: searching for code review, the first two results had zero votes and one vote.
Ask your agent to pull at least ten results instead of one and to choose by looking at the vote count. If you know what kind of thing you want, give it a category or tag filter.
Anyone can add prompts and skills, and content is not reviewed up front. Skills can contain script files too, and your agent can pull all of them with get_skill.
Read a skill's files before installing it, especially the scripts. When your agent wants to install one, tell it to show you the contents first.
You can tell your agent this in one sentence. Something like this works:
Search prompts.chat for "landing page copy" with 20 results,
show me the three with the most votes and tell me why you picked each.When you need an account
Searching and pulling prompts needs no account or key, it works the moment you install it. If you want to save your own prompts or skills to the library, you create an account on prompts.chat and get an API key from the settings page.
Do not put the key in the address
The server accepts the key both as a header and at the end of the address. A key written into the address ends up in logs and config files as plain text. Pass it as a header, or if you use the plugin, put it in the PROMPTS_API_KEY environment variable.
claude mcp add --transport http prompts.chat https://prompts.chat/api/mcp --header "PROMPTS_API_KEY: your-key"How big it really is
I asked the site's own API on 24 September 2026: the public listing holds 2,372 entries. Of those, 1,753 are text, 477 image, 54 video and 77 agent skills. Figures above 10,000 get passed around; counting private and delisted prompts that may be true, but this is the part you or your agent can actually reach.
That is not small. As a starting point for writing, code and images it is more than enough, and all of it is free. Just do not trust your agent blindly; have it check the votes as above, and read skills before you install them.
If you want a private copy for your own team, the project runs on your own server too. One command scaffolds a new library, and the rest is in the repo's own setup guide.
npx prompts.chat new my-prompt-libraryDOA: Yapay Zeka ve Otomasyon
Installing these tools on your own is one thing; actually building with them is another. The community has people using these daily and people building systems from scratch.
If you want a system that actually runs in your business, let's talk for 10 minutes; I'll look at what you're trying to build and tell you which path fits. Free, and not a sales pitch.
Book a 10-minute callIf you'd rather learn this alongside people doing the same work instead of on your own, the community is always open:
Join the communityThis is an affiliate link.