@ismalicious/mcp-server

@ismalicious/mcp-server

An MCP server for the isinjected gate. It gives an AI agent two tools to call before it acts on untrusted content:

Tool	What it does
scan_before_use	Scans a block of content for prompt injection and checks any URLs/domains/IPs it contains against ~30M threat indicators. Returns block | warn | allow.
check_url	Reputation-only pre-fetch gate for a single URL/domain/IP.

Zero runtime dependencies — a plain JSON-RPC 2.0 server over stdio.

Setup

Get an API key/secret at https://ismalicious.com/app/account, then add the server to your MCP client config:

{
  "mcpServers": {
    "isinjected": {
      "command": "npx",
      "args": ["-y", "@ismalicious/mcp-server"],
      "env": {
        "ISMALICIOUS_API_KEY": "your-api-key",
        "ISMALICIOUS_API_SECRET": "your-api-secret"
      }
    }
  }
}

ISMALICIOUS_API_BASE is optional and defaults to https://ismalicious.com/api.

Free tier

Every plan includes a monthly scan allowance (Free: 1,000 scans/month). Scans are metered separately from API request quota. See https://ismalicious.com/pricing.

Example

An agent that fetches a web page should scan it before summarizing:

scan_before_use({ content: "<page text>", source_url: "https://example.com" })
→ { "verdict": "block", "injection": { "score": 0.94, "families": ["instruction_override"] }, ... }

A block verdict means the content tried to hijack the agent — do not act on it.