MCP Prompt Collector

An MCP server that provides interactive tools for logging, analyzing, and improving your Claude Desktop prompts to help you develop better prompt engineering skills.

Installation

Option 1: NPM Package (Recommended)

# Install globally
npm install -g mcp-prompt-collector

# Or install locally in your project
npm install mcp-prompt-collector

Option 2: From Source

git clone https://github.com/daneb/improver-mcp.git
cd improver-mcp
npm install
npm run build

Setup

1. Configure Claude Desktop

   Add this to your ~/Library/Application Support/Claude/claude_desktop_config.json:

   For global npm install:

   {
     "mcpServers": {
       "mcp-prompt-collector": {
         "command": "mcp-prompt-collector",
         "env": {
           "HOME": "/Users/daneb"
         }
       }
     }
   }
   

   For source installation:

   {
     "mcpServers": {
       "mcp-prompt-collector": {
         "command": "/path/to/your/node",
         "args": ["/path/to/improver-mcp/dist/index.js"],
         "env": {
           "HOME": "/Users/daneb"
         }
       }
     }
   }
   

2. Restart Claude Desktop to load the MCP server

3. Access the dashboard at http://localhost:3456

What It Does

• 🛠️ Interactive Tools: Provides 4 tools accessible in Claude Desktop for prompt analysis
• 📊 Quality Analysis: Scores prompts on structure, clarity, context, and effectiveness
• 💡 Smart Insights: Identifies patterns and suggests specific improvements
• 📈 Progress Tracking: Shows quality trends and improvement over time
• 🌐 Web Dashboard: Clean interface to view history, analytics, and insights

Features

Core Functionality

• ✅ Interactive MCP tools for Claude Desktop integration
• ✅ SQLite database for local storage (no cloud sync)
• ✅ Real-time prompt quality scoring
• ✅ Web dashboard with analytics
• ✅ Automated insights generation

Available Tools

• log_prompt - Log and analyze a prompt with immediate quality feedback
• improve_prompt - Get specific suggestions for improving any prompt
• get_prompt_insights - View analytics and patterns from your prompt history
• view_stats - Display comprehensive statistics and quality trends

Analysis Engine

• Quality Scoring: 0-10 score based on structure, clarity, and completeness
• Complexity Detection: Categorizes prompts as simple, moderate, or complex
• Technique Suggestions: Recommends optimal prompting techniques
• Issue Detection: Identifies ambiguity, missing context, bias, etc.

Dashboard Features

• Prompt history with search and filtering
• Quality trends over time
• Complexity distribution charts
• Recent activity overview
• Actionable improvement insights

Usage

In Claude Desktop

Once the MCP server is running, you'll have access to these tools in Claude Desktop:

log_prompt
- Use: Log and analyze any prompt
- Example: "log_prompt" with your prompt text
- Returns: Quality score, complexity, and improvement suggestions

improve_prompt  
- Use: Get specific suggestions for a prompt
- Example: "improve_prompt" with your prompt and optional goal
- Returns: Detailed analysis and actionable improvement recommendations

get_prompt_insights
- Use: View patterns and insights from your prompt history
- Example: "get_prompt_insights" optionally filtered by category
- Returns: Analytics on your prompting patterns and common issues

view_stats
- Use: See comprehensive statistics and trends
- Example: "view_stats" for last 30 days (configurable)
- Returns: Total prompts, quality trends, complexity breakdown

Development Commands

# Build and run
npm run build                       # Build TypeScript
npm run dev                         # Development mode with auto-reload
npm start                          # Start the MCP server

# Code quality
npm test                           # Run tests
npm run lint                       # Check code style
npm run format                     # Format code

How It Works

1. MCP Integration: Registers as an MCP server with Claude Desktop
2. Interactive Tools: Provides 4 tools accessible through Claude Desktop interface
3. Manual Logging: Users explicitly log prompts using the log_prompt tool
4. Analysis: Scores quality, detects patterns, identifies improvements
5. Storage: Saves everything locally in SQLite database
6. Insights: Generates personalized suggestions based on your patterns
7. Dashboard: Provides web interface to view analytics and history

Configuration

Default configuration is stored in ~/.mcp-prompt-collector/config/settings.json:

{
  "name": "mcp-prompt-collector",
  "capabilities": {
    "intercept": true,
    "modify": false,
    "store": true
  },
  "settings": {
    "enhancePrompts": false,
    "storageLocation": "~/.mcp-prompt-collector/data",
    "dashboardPort": 3456,
    "analysisLevel": "detailed",
    "retentionDays": 90
  }
}

Data Storage

All data is stored locally in SQLite database:

• Location: ~/.mcp-prompt-collector/data/prompt_history.db
• Privacy: No cloud sync, complete local control
• Retention: Configurable cleanup after N days
• Export: Full data export capabilities

Development

# Clone and setup
git clone <repo-url>
cd mcp-prompt-collector
npm install

# Development workflow
npm run dev        # Watch mode
npm run build      # Production build
npm test          # Run tests
npm run lint      # Code quality

# Testing
npm run test:unit        # Unit tests
npm run test:integration # Integration tests

Architecture

• MCP Protocol Handler: Communicates with Claude Desktop
• Storage Manager: SQLite database operations
• Analysis Engine: Prompt quality scoring and insights
• Dashboard Server: Express.js web interface
• Background Worker: Scheduled analysis and cleanup

Privacy & Security

• ✅ Local Only: All data stays on your machine
• ✅ No Cloud Sync: Zero external data transmission
• ✅ User Control: Complete data ownership and export
• ✅ Transparent: Open source, auditable code

Requirements

• Node.js: 18+
• Claude Desktop: Latest version with MCP support
• macOS: Primary support (Windows/Linux coming soon)

Troubleshooting

Server won't start

• Check if port 3456 is available
• Verify Claude Desktop MCP configuration
• Run mcp-prompt-collector status for diagnostics

No prompts showing in dashboard

• Make sure you're using the log_prompt tool in Claude Desktop
• Prompts are not automatically captured - you must manually log them
• Check that the MCP server tools are available in Claude Desktop

Dashboard not loading

• Check if server is running on correct port
• Try different port with setup --port 3457
• Check browser console for errors

License

MIT License - see LICENSE file for details