What are prompts.chat and MCP? Finding prompts for AI
Community prompts can help you get started, but you still need to read their instructions and adapt them to the task. prompts.chat offers a website for manual browsing and MCP access for agents to search while working. This article explains what each route provides and when an API key or an additional service is needed.

prompts.chat is a library; MCP connects its tools
prompts.chat evolved from Awesome ChatGPT Prompts to support finding, sharing, and storing prompts. Users can select text on the website and adapt it themselves. A library does not establish that every prompt has been tested with every model or language.
MCP connects tools to AI applications. Here, the server lets an Agent search for and retrieve prompts without manual copying. The client must support MCP, and the user still controls tool permissions.
Searching, retrieving, and saving are different actions
The plugin documentation lists search_prompts for searching by keyword, category, tag, or type, and get_prompt for retrieval with variable substitution. save_prompt writes data and requires an API key. Permission to search and read should not be treated as permission to publish or save new text.
It also provides search_skills, get_skill and tools for editing Skill files. A single server can expose both read and write actions. Check the list in the plugin documentation before enabling it.
Choose the website, plugin, or MCP for the task
For occasional searches, the website may be enough. If the agent needs to search while it works, the README lists https://prompts.chat/api/mcp and this command to start a local MCP server.
npx -y prompts.chat mcp
The plugin also supplies prompt and Skill search commands in Claude Code.
A locally running server does not guarantee local processing for every function; it may still call remote services. We reviewed documentation without adding the server or executing MCP searches, so we have no experimental measurements of speed or search quality.
Adapting a code-review prompt
Suppose you find a code-review prompt. Add the framework version, target files, expected behavior, and required evidence: for example, “Report problems with an explainable failure path, including file, location, impact, and reproduction steps.” Remove instructions outside the task's scope.
If you only want a review but the prompt orders a complete reimplementation, revise it before use. This is KrobClick's suggested approach, not an actual search result or a claim that a named, scored prompt exists.
Check instructions, rights, and data flows
The project separates its code from prompt data. It identifies code and site-authored content as MIT-licensed and user-submitted prompt data as CC0. Check the actual files before republishing images, text, or code rather than relying on the general label “open source.”
Community prompts still need reading: they can contain overly broad instructions or requests to disclose information. Do not include secrets or customer data in searches or prompt-improvement requests without checking the receiving service.
Save a working prompt with checks for its output
Store the adapted prompt with example input, expected output, and known failure cases. Test normal and edge cases, and recheck after changing models before important work. The size of a library does not identify the prompt that fits your task.
If you find a Skill and want to install it, see How to inspect a Skill before installation to assess permissions and bundled files.
Sources
- f/prompts.chat: Documentation for the referenced version
- prompts.chat: Plugin and MCP tools
- prompts.chat: Dual licensing
- f/prompts.chat
Reviewed f/prompts.chat commit 0d2f9b3b3b1f31cd787cea3ce210d9c229229c09 on October 30, 2026, Thailand time. This article explains upstream documentation; we did not install or test a live service connection. Capabilities and terms may vary by version.