Prompts MCP Server

A Model Context Protocol (MCP) server for managing and providing prompts. This server allows users and LLMs to easily add, retrieve, and manage prompt templates stored as markdown files with YAML frontmatter support.

Quick Start

# 1. Install
git clone https://github.com/tanker327/prompts-mcp-server.git
cd prompts-mcp-server
npm install && npm run build

# 2. Test it works
npm test

# 3. Add to your MCP client config (e.g., Claude Desktop)
# Add this to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
  "mcpServers": {
    "prompts-mcp-server": {
      "command": "node",
      "args": ["/path/to/prompts-mcp-server/dist/index.js"]
    }
  }
}

# 4. Restart your MCP client and start using the tools!

Features

Add Prompts: Store new prompts as markdown files with YAML frontmatter
Retrieve Prompts: Get specific prompts by name
List Prompts: View all available prompts with metadata preview
Delete Prompts: Remove prompts from the collection
File-based Storage: Prompts are stored as markdown files in the prompts/ directory
Real-time Caching: In-memory cache with automatic file change monitoring
YAML Frontmatter: Support for structured metadata (title, description, tags, etc.)
TypeScript: Full TypeScript implementation with comprehensive type definitions
Modular Architecture: Clean separation of concerns with dependency injection
Comprehensive Testing: 95 tests with 84.53% code coverage

Installation

Option 1: From GitHub (Recommended)

# Clone the repository
git clone https://github.com/tanker327/prompts-mcp-server.git
cd prompts-mcp-server

# Install dependencies
npm install

# Build the TypeScript code
npm run build

# Test the installation
npm test

Option 2: Direct Download

Download the latest release from GitHub
Extract to your desired location
Run installation steps above

Option 3: Development Setup

# Clone and set up for development
git clone https://github.com/tanker327/prompts-mcp-server.git
cd prompts-mcp-server
npm install

# Start development server with auto-reload
npm run dev

# Run tests in watch mode
npm run test:watch

Verification

After installation, verify the server works:

# Start the server (should show no errors)
npm start

# Or test with MCP Inspector
npx @modelcontextprotocol/inspector node dist/index.js

Testing

Run the comprehensive test suite:

npm test

Run tests with coverage:

npm run test:coverage

Watch mode for development:

npm run test:watch

MCP Tools

The server provides the following tools:

`add_prompt`

Add a new prompt to the collection. If no YAML frontmatter is provided, default metadata will be automatically added.

name (string): Name of the prompt
content (string): Content of the prompt in markdown format with optional YAML frontmatter

`create_structured_prompt`

Create a new prompt with guided metadata structure and validation.

name (string): Name of the prompt
title (string): Human-readable title for the prompt
description (string): Brief description of what the prompt does
category (string, optional): Category (defaults to "general")
tags (array, optional): Array of tags for categorization (defaults to ["general"])
difficulty (string, optional): "beginner", "intermediate", or "advanced" (defaults to "beginner")
author (string, optional): Author of the prompt (defaults to "User")
content (string): The actual prompt content (markdown)

`get_prompt`

Retrieve a prompt by name.

name (string): Name of the prompt to retrieve

`list_prompts`

List all available prompts with metadata preview. No parameters required.

`delete_prompt`

Delete a prompt by name.

name (string): Name of the prompt to delete

Usage Examples

Once connected to an MCP client, you can use the tools like this:

Method 1: Quick prompt creation with automatic metadata

// Add a prompt without frontmatter - metadata will be added automatically
add_prompt({
  name: "debug_helper",
  content: `# Debug Helper

Help me debug this issue by:
1. Analyzing the error message
2. Suggesting potential causes
3. Recommending debugging steps`
})
// This automatically adds default frontmatter with title "Debug Helper", category "general", etc.

Method 2: Structured prompt creation with full metadata control

// Create a prompt with explicit metadata using the structured tool
create_structured_prompt({
  name: "code_review",
  title: "Code Review Assistant",
  description: "Helps review code for best practices and potential issues",
  category: "development",
  tags: ["code", "review", "quality"],
  difficulty: "intermediate",
  author: "Development Team",
  content: `# Code Review Prompt

Please review the following code for:
- Code quality and best practices
- Potential bugs or issues
- Performance considerations
- Security vulnerabilities

## Code to Review
[Insert code here]`
})

Method 3: Manual frontmatter (preserves existing metadata)

// Add a prompt with existing frontmatter - no changes made
add_prompt({
  name: "custom_prompt",
  content: `---
title: "Custom Assistant"
category: "specialized"
tags: ["custom", "specific"]
difficulty: "advanced"
---

# Custom Prompt Content
Your specific prompt here...`
})

Other operations

// Get a prompt
get_prompt({ name: "code_review" })

// List all prompts (shows metadata preview)
list_prompts({})

// Delete a prompt
delete_prompt({ name: "old_prompt" })

File Structure

prompts-mcp-server/
├── src/
│   ├── index.ts          # Main server orchestration
│   ├── types.ts          # TypeScript type definitions
│   ├── cache.ts          # Caching system with file watching
│   ├── fileOperations.ts # File I/O operations
│   └── tools.ts          # MCP tool definitions and handlers
├── tests/
│   ├── helpers/
│   │   ├── testUtils.ts  # Test utilities
│   │   └── mocks.ts      # Mock implementations
│   ├── cache.test.ts     # Cache module tests
│   ├── fileOperations.test.ts # File operations tests
│   ├── tools.test.ts     # Tools module tests
│   └── index.test.ts     # Integration tests
├── prompts/              # Directory for storing prompt markdown files
│   ├── code_review.md
│   ├── debugging_assistant.md
│   └── api_design.md
├── dist/                 # Compiled JavaScript output
├── CLAUDE.md            # Development documentation
├── package.json
├── tsconfig.json
└── README.md

Architecture

The server uses a modular architecture with the following components:

PromptCache: In-memory caching with real-time file change monitoring via chokidar
PromptFileOperations: File I/O operations with cache integration
PromptTools: MCP tool definitions and request handlers
Type System: Comprehensive TypeScript types for all data structures

YAML Frontmatter Support

Prompts can include structured metadata using YAML frontmatter:

---
title: "Prompt Title"
description: "Brief description of the prompt"
category: "development"
tags: ["tag1", "tag2", "tag3"]
difficulty: "beginner" | "intermediate" | "advanced"
author: "Author Name"
version: "1.0"
---

# Prompt Content

Your prompt content goes here...

MCP Client Configuration

This server can be configured with various MCP-compatible applications. Here are setup instructions for popular clients:

Claude Desktop

Add this to your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "prompts-mcp-server": {
      "command": "node",
      "args": ["/path/to/prompts-mcp-server/dist/index.js"],
      "env": {
        "PROMPTS_DIR": "/path/to/your/prompts/directory"
      }
    }
  }
}

Cline (VS Code Extension)

Add to your Cline MCP settings in VS Code:

{
  "cline.mcp.servers": {
    "prompts-mcp-server": {
      "command": "node",
      "args": ["/path/to/prompts-mcp-server/dist/index.js"],
      "env": {
        "PROMPTS_DIR": "/path/to/your/prompts/directory"
      }
    }
  }
}

Continue.dev

In your ~/.continue/config.json:

{
  "mcpServers": [
    {
      "name": "prompts-mcp-server",
      "command": "node",
      "args": ["/path/to/prompts-mcp-server/dist/index.js"],
      "env": {
        "PROMPTS_DIR": "/path/to/your/prompts/directory"
      }
    }
  ]
}

Zed Editor

In your Zed settings (~/.config/zed/settings.json):

{
  "assistant": {
    "mcp_servers": {
      "prompts-mcp-server": {
        "command": "node",
        "args": ["/path/to/prompts-mcp-server/dist/index.js"],
        "env": {
          "PROMPTS_DIR": "/path/to/your/prompts/directory"
        }
      }
    }
  }
}

Custom MCP Client

For any MCP-compatible application, use these connection details:

Protocol: Model Context Protocol (MCP)
Transport: stdio
Command: node /path/to/prompts-mcp-server/dist/index.js
Environment Variables:
- PROMPTS_DIR: Custom directory for storing prompts (optional, defaults to ./prompts)

Development/Testing Setup

For development or testing with the MCP Inspector:

# Install MCP Inspector
npm install -g @modelcontextprotocol/inspector

# Run the server with inspector
npx @modelcontextprotocol/inspector node dist/index.js

Docker Configuration

Create a docker-compose.yml for containerized deployment:

version: '3.8'
services:
  prompts-mcp-server:
    build: .
    environment:
      - PROMPTS_DIR=/app/prompts
    volumes:
      - ./prompts:/app/prompts
    stdin_open: true
    tty: true

Server Configuration

The server automatically creates the prompts/ directory if it doesn't exist
Prompt files are automatically sanitized to use safe filenames (alphanumeric characters, hyphens, and underscores only)
File changes are monitored in real-time and cache is updated automatically
Prompts directory can be customized via the PROMPTS_DIR environment variable

Environment Variables

Variable	Description	Default
`PROMPTS_DIR`	Directory to store prompt files	`./prompts`
`NODE_ENV`	Environment mode	`production`

Requirements

Node.js 18.0.0 or higher
TypeScript 5.0.0 or higher
Dependencies:
- @modelcontextprotocol/sdk ^1.0.0
- gray-matter ^4.0.3 (YAML frontmatter parsing)
- chokidar ^3.5.3 (file watching)

Development

The project includes comprehensive tooling for development:

TypeScript: Strict type checking and modern ES modules
Vitest: Fast testing framework with 95 tests and 84.53% coverage
ESLint: Code linting (if configured)
File Watching: Real-time cache updates during development

Troubleshooting

Common Issues

"Module not found" errors

# Ensure TypeScript is built
npm run build

# Check that dist/ directory exists and contains .js files
ls dist/

MCP client can't connect

Verify the server starts without errors: npm start
Check the correct path is used in client configuration
Ensure Node.js 18+ is installed: node --version
Test with MCP Inspector: npx @modelcontextprotocol/inspector node dist/index.js

Permission errors with prompts directory

# Ensure the prompts directory is writable
mkdir -p ./prompts
chmod 755 ./prompts

File watching not working

On Linux: Install inotify-tools
On macOS: No additional setup needed
On Windows: Ensure Windows Subsystem for Linux (WSL) or native Node.js

Debug Mode

Enable debug logging by setting environment variables:

# Enable debug mode
DEBUG=* node dist/index.js

# Or with specific debug namespace
DEBUG=prompts-mcp:* node dist/index.js

Getting Help

Check the GitHub Issues
Review the test files for usage examples
Use MCP Inspector for debugging client connections
Check your MCP client's documentation for configuration details

Performance Tips

The server uses in-memory caching for fast prompt retrieval
File watching automatically updates the cache when files change
Large prompt collections (1000+ files) work efficiently due to caching
Consider using SSD storage for better file I/O performance

License

MIT