Prompt Enhancement MCP Server

An MCP (Model Context Protocol) server that enhances user prompts using AI. It takes a rough prompt and returns a more detailed, clear, and effective version. Works with any MCP-compatible client like Claude Desktop or Claude Code.

Quick Start

1. Install and set your API key:

   export ANTHROPIC_API_KEY=sk-ant-...
   

2. Add to your MCP client config (e.g. Claude Desktop):

   {
     "mcpServers": {
       "prompt-enhancer": {
         "command": "npx",
         "args": ["-y", "prompt-enhancement-mcp-server"],
         "env": {
           "ANTHROPIC_API_KEY": "your-key-here"
         }
       }
     }
   }
   

3. Use the enhance_prompt tool to improve any prompt.

Installation

# Run directly with npx (recommended)
npx -y prompt-enhancement-mcp-server

# Or install globally
npm install -g prompt-enhancement-mcp-server

# Or install locally
npm install prompt-enhancement-mcp-server

Supported Providers

Set at least one API key as an environment variable.

Configuration

Environment Variables

# API keys (set at least one)
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
OPENROUTER_API_KEY=sk-or-...
GEMINI_API_KEY=...

# Optional settings
PROMPT_ENHANCER_DEFAULT_PROVIDER=anthropic   # Which provider to use by default
PROMPT_ENHANCER_CONFIG=/path/to/config.json  # Custom config file path
PROMPT_ENHANCER_LOG_LEVEL=info               # debug, info, warn, error

Optional Config File

Create ~/.config/prompt-enhancer/config.json for model defaults, custom templates, or OpenAI-compatible endpoints:

{
  "defaultProvider": "anthropic",
  "providers": {
    "anthropic": {
      "model": "claude-sonnet-4-5-20250929",
      "temperature": 0.7,
      "maxTokens": 4096
    },
    "openai": {
      "model": "gpt-4o"
    },
    "openrouter": {
      "model": "anthropic/claude-sonnet-4-5-20250929",
      "baseUrl": "https://openrouter.ai/api/v1"
    },
    "gemini": {
      "model": "gemini-2.0-flash"
    }
  },
  "templates": {
    "default": "Generate an enhanced version of this prompt (reply with only the enhanced prompt - no conversation, explanations, lead-in, bullet points, placeholders, or surrounding quotes):\n\n${userInput}"
  },
  "options": {
    "maxContextMessages": 10,
    "contextTruncateLength": 500
  }
}

API keys are never stored in the config file -- always use environment variables.

Tool: enhance_prompt

The server exposes a single MCP tool:

Claude Desktop Integration

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "prompt-enhancer": {
      "command": "npx",
      "args": ["-y", "prompt-enhancement-mcp-server"],
      "env": {
        "ANTHROPIC_API_KEY": "your-key-here"
      }
    }
  }
}

Claude Code Integration

Add to ~/.claude/claude_code_config.json or your project's .mcp.json:

{
  "mcpServers": {
    "prompt-enhancer": {
      "command": "npx",
      "args": ["-y", "prompt-enhancement-mcp-server"],
      "env": {
        "OPENAI_API_KEY": "your-key-here",
        "PROMPT_ENHANCER_DEFAULT_PROVIDER": "openai"
      }
    }
  }
}

Custom Templates

Override the default enhancement template via the config file or the template tool parameter:

{
  "templates": {
    "default": "Rewrite this prompt to be more specific and actionable:\n\n${userInput}"
  }
}

Or pass inline when calling the tool:

{
  "text": "write a web app",
  "template": "Add technical details and edge cases to this prompt:\n\n${userInput}"
}

OpenAI-Compatible Endpoints

Connect to Ollama, LM Studio, vLLM, or any OpenAI-compatible API:

{
  "providers": {
    "openai-compatible": {
      "baseUrl": "http://localhost:11434/v1",
      "model": "llama3"
    }
  }
}

Then use it:

{
  "text": "write a function",
  "provider": "openai-compatible"
}

Troubleshooting

"Unsupported provider" error Check that the provider name is one of: anthropic, openai, openrouter, gemini, openai-compatible.

"API key is required" / authentication errors Make sure the corresponding environment variable is set. API keys are loaded from env vars, not the config file.

Timeouts or rate limit errors The server retries transient errors up to 3 times with exponential backoff. If you're consistently hitting rate limits, try a different provider or model.

No output / server won't start Check PROMPT_ENHANCER_LOG_LEVEL=debug for detailed logging. Logs go to stderr (MCP protocol reserves stdout for communication).

Development

git clone <repo-url>
cd prompt-enhancement-mcp-server
npm install
npm run build
npm test

License

Apache-2.0