Prompt Defend MCP Server (Python)

A Model Context Protocol (MCP) server that exposes Prompt Defend's 16-layer guardrail system as tools for AI agents.

Installation

pip install promptdefend-mcp

Or install from source:

cd framework/promptdefend-mcp/python
pip install -e .

Quick Start

As a Standalone Server

promptdefend-mcp --api-key your-api-key

With Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "promptdefend": {
      "command": "promptdefend-mcp",
      "args": ["--api-key", "your-api-key"]
    }
  }
}

Available Tools

scan_prompt

Scan a prompt for security threats using 16-layer guardrails.

Input:

{
  "prompt": "string - The prompt to scan"
}

Output:

{
  "safe": true,
  "reason": "No threats detected",
  "details": {...}
}

check_blocklist

Check if a prompt contains known attack keywords.

Input:

{
  "prompt": "string - The prompt to check"
}

check_allowlist

Check if a prompt matches safe/educational phrases.

Input:

{
  "prompt": "string - The prompt to check"
}

Environment Variables

License

Proprietary - See LICENSE for details.

© 2026 Prompt Defend. All Rights Reserved.