cisco-ai-defense/mcp-scanner

Scan MCP servers for potential threats & security findings.

What it solves

MCP Scanner addresses the security risks associated with the Model Context Protocol (MCP), specifically the danger of malicious MCP servers, tools, prompts, and resources. It helps developers and security researchers detect threats like tool poisoning, prompt injection, and malware bundled with MCP servers.

How it works

The tool employs a multi-engine approach to security analysis:

  • Scanning Engines: It combines the Cisco AI Defense inspect API, YARA rules for pattern matching, and an "LLM-as-a-judge" system for semantic analysis.
  • Broad Coverage: It scans various components including MCP tools, prompts, resources, server instructions, and the source code of the servers themselves.
  • Additional Checks: It integrates pip-audit for Python dependency vulnerabilities, VirusTotal hash lookups for binary malware detection, and Docker-sandboxed behavioral analysis for PyPI and npm packages.
  • Deployment: It can be used as a standalone CLI tool, a REST API server, or integrated as a Python SDK.

Who it’s for

It is designed for security engineers, AI developers, and organizations deploying MCP-based AI agents who need to verify the safety and integrity of the tools and servers they connect to.

Highlights

  • Multi-Engine Analysis: Leverages YARA, LLMs, and Cisco AI Defense for comprehensive detection.
  • Flexible Connectivity: Supports scanning remote servers (SSE/HTTP), stdio servers, and offline JSON files for CI/CD pipelines.
  • Behavioral Analysis: Detects mismatches between what a tool's documentation claims it does and what the code actually implements.
  • Supply Chain Security: Scans PyPI and npm packages in sandboxes and checks for known CVEs in dependencies.