MCP Integration - Prompt Inspector

What is MCP?

The Model Context Protocol (MCP) is an open standard that allows AI applications to connect with external tools and services. Prompt Inspector provides an MCP server that exposes its detection capabilities as a tool, enabling AI coding assistants and agents to check user inputs for prompt injection attacks in real time.

Supported Clients

The Prompt Inspector MCP server works with any MCP-compatible client, including:

Cursor — AI-powered code editor
VS Code Copilot — GitHub Copilot Chat in VS Code
Claude Desktop — Anthropic’s Claude desktop app
Dify — LLM app development platform
Any SSE-compatible MCP client

How to Connect

Authentication

All MCP clients must provide your App API Key when connecting. The key can be passed via:

X-App-Key header (recommended)
Authorization: Bearer <key> header

Cursor

Add the following to your Cursor MCP settings (.cursor/mcp.json):

{
  "mcpServers": {
    "prompt-inspector": {
      "url": "https://promptinpector.io/sse",
      "headers": {
        "X-App-Key": "your-api-key"
      }
    }
  }
}

VS Code (Copilot Chat)

Add to your VS Code settings.json:

{
  "mcp": {
    "servers": {
      "prompt-inspector": {
        "type": "sse",
        "url": "https://promptinpector.io/sse",
        "headers": {
          "X-App-Key": "your-api-key"
        }
      }
    }
  }
}

Claude Desktop

Add to your Claude Desktop configuration file (claude_desktop_config.json):

{
  "mcpServers": {
    "prompt-inspector": {
      "url": "https://promptinpector.io/sse",
      "headers": {
        "X-App-Key": "your-api-key"
      }
    }
  }
}

Dify

In the Dify platform, add the MCP server as a tool provider:

Go to Tools → Custom Tools
Add a new MCP tool with the SSE endpoint URL: https://promptinpector.io/sse
Configure the X-App-Key header with your API key

Available Tools

The MCP server exposes one tool:

`detect`

Analyzes text for prompt injection attacks and malicious content. Parameters:

Parameter	Type	Required	Description
`text`	`string`	Yes	The text to analyze for prompt injection

Returns: A human-readable string containing:

Safety status (SAFE / THREAT DETECTED)
Risk score (0.0 = safe, 1.0 = certain threat)
Detected threat categories
Processing latency and request ID
Raw JSON for programmatic parsing

Example output (safe input):

✅ Detection result: SAFE
Risk score : N/A (0.0 = safe, 1.0 = certain threat)
Categories : none
Latency    : 15 ms
Request ID : abc-123-def-456

Example output (threat detected):

🚨 Detection result: THREAT DETECTED
Risk score : 0.95 (0.0 = safe, 1.0 = certain threat)
Categories : prompt_injection
Latency    : 42 ms
Request ID : def-789-ghi-012

Raw JSON   : {"request_id": "def-789-ghi-012", "is_safe": false, "score": 0.95, "category": ["prompt_injection"], "latency_ms": 42}

View Source on GitHub

Find the complete MCP server source code and setup instructions in the repository.

​What is MCP?

​Supported Clients

​How to Connect

​Authentication

​Cursor

​VS Code (Copilot Chat)

​Claude Desktop

​Dify

​Available Tools

​detect