MCP Prompt Collector

MCP Prompt Collector is an MCP server that provides tools for logging, analyzing, and improving prompts directly within Claude Desktop. Designed for developers, prompt engineers, and AI practitioners, it bridges Claude with a local analysis engine and an Express-powered web dashboard. Instead of sending prompt histories to external cloud services, the server stores all metrics, complexity breakdowns, and scoring data locally inside an SQLite database. Users can inspect their prompt writing habits, detect common issues such as ambiguity or missing context, and receive immediate quality ratings scored from 0 to 10. The system also tracks quality trends over time, allowing individuals to review historical usage through a dedicated browser dashboard hosted on their machine. By exposing tools to log inputs and request targeted prompt revisions, the server helps users refine their prompting techniques systematically across sessions while maintaining complete data ownership.

Category: AI & LLM Tooling

Tags: claude, optimization, prompt-engineering, prompt-management

Visit MCP Prompt Collector

How to install and configure MCP Prompt Collector

1. Install the package globally via npm: bash npm install -g mcp-prompt-collector Alternatively, clone the repository, run npm install, and build with npm run build. 2. Open your Claude Desktop configuration file at ~/Library/Application Support/Claude/claude_desktop_config.json and add the server definition: json { "mcpServers": { "mcp-prompt-collector": { "command": "mcp-prompt-collector", "env": { "HOME": "/Users/your-username" } } } } (Replace /Users/your-username with your home directory path, or point to your node binary and built dist/index.js if installing from source). 3. Restart Claude Desktop to apply the configuration. 4. Access the web dashboard locally at http://localhost:3456 to view your analytics.

What you can do with MCP Prompt Collector

• Log and evaluate prompt quality in real time using the log_prompt tool to receive numerical scores and suggestions. - Request targeted revisions and technique advice for complex instructions using the improve_prompt tool inside Claude Desktop. - Track historical prompt quality trends, complexity distributions, and aggregate metrics over time using the view_stats tool. - Identify recurring patterns, missing contextual details, and prompt ambiguity across sessions via the get_prompt_insights command. - Search and review previous prompt records via the local web dashboard running at localhost without cloud dependencies.

Key facts

• https://github.com/daneb/improver-mcp
• AI & LLM Tooling, Developer Tools & Code Intelligence
• claude, optimization, prompt-engineering, prompt-management

Part of MCP Servers

How do I install MCP Prompt Collector?

You can install MCP Prompt Collector by running npm install -g mcp-prompt-collector. After installing, register it under the mcpServers key in your Claude Desktop configuration file and restart Claude Desktop to load the tools.

Which MCP clients work with MCP Prompt Collector?

MCP Prompt Collector is primarily designed to work with Claude Desktop on macOS, using Node.js version 18 or higher. It interacts with Claude through standard Model Context Protocol tool declarations.

What tools are provided by MCP Prompt Collector?

The server exposes four primary tools: log_prompt to record and score prompt quality, improve_prompt to get concrete optimization suggestions, get_prompt_insights to inspect historical usage patterns, and view_stats to view overall prompt performance statistics.

Does MCP Prompt Collector automatically capture all my prompts?

No. The server does not automatically intercept your conversations. You must explicitly log prompts using the log_prompt tool inside Claude Desktop to record and analyze them in your local database.

Is MCP Prompt Collector open source?

Yes, MCP Prompt Collector is open source and licensed under the MIT License. The complete source code and configuration details are available on GitHub.