prompts

Prompts MCP Server provides a lightweight, file-backed catalog for creating, storing, and retrieving Markdown prompts with optional YAML frontmatter. It exposes a small set of MCP tools to manage your prompt templates: add_prompt to store new prompts (with optional frontmatter that will be enriched if missing), create_structured_prompt to create prompts with explicit metadata such as title, description, category, tags, and difficulty, get_prompt to fetch a specific prompt by name, list_prompts to view all prompts with metadata previews, and delete_prompt to remove prompts by name. Prompts are kept in the prompts/ directory and benefit from a real-time in-memory cache that updates when files change on disk, ensuring clients always see current data without manual reloads.

To use it, install the server globally (or run via your MCP client integration) and configure your MCP client to point to the prompts-mcp-server. Once connected, you can add new prompts by name and content, retrieve prompts for inclusion in your workflows, list available prompts with their metadata, or delete prompts you no longer need. YAML frontmatter support enables structured metadata (title, description, category, tags, difficulty, author, version) to accompany your Markdown prompts, making it easy for LLMs and clients to discover and categorize prompts.

Prerequisites:

• Node.js and npm installed on your system
• Internet access to install packages from npm registry

Option A: Install globally from npm (recommended for simple usage)

1. Install the server globally: npm install -g prompts-mcp-server
2. Start the server (the CLI provides the prompts-mcp-server command): prompts-mcp-server
3. Configure your MCP client to use the server name prompts-mcp-server and the command as shown in your MCP client config.

Option B: Development or local usage (no global install)

1. Clone the repository (or obtain the package locally)
2. Install dependencies: npm install
3. Build if needed (TypeScript projects): npm run build
4. Start the server (if the script is defined in package.json): npm start

Option C: Verification

• After installation, verify the server works by starting it and, if available, testing with an MCP inspector tool: npx @modelcontextprotocol/inspector prompts-mcp-server

Tips and common considerations:

• Prompts are stored as Markdown files in the prompts/ directory and can include YAML frontmatter for metadata such as title, description, category, tags, and difficulty.
• Real-time caching uses a file-watching mechanism to keep in-memory data in sync with on-disk prompts.
• Ensure your MCP client configuration uses the correct server name (e.g., prompts-mcp-server) and points to the correct command if you are running via a custom setup.
• If you enable structured prompts with create_structured_prompt, you can predefine metadata fields to improve searchability and organization.
• When running in environments with restricted file system access, verify that the prompts/ directory is writable by the server process.
• If you encounter issues with frontmatter parsing, verify that your Markdown files begin with a proper YAML frontmatter block and that blocks are terminated correctly with --- markers.