MCP.so
ログイン
サーバー

MCP-Scan: An MCP Security Scanner

@invariantlabs-ai

Security scanner for AI agents, MCP servers and agent skills.

概要

What is MCP-Scan: An MCP Security Scanner?

MCP-Scan (also known as Snyk Agent Scan) is a security scanner that discovers and scans agent components on your machine—including MCP servers, agent tools, and skills—for prompt injections, malware payloads, and other vulnerabilities. It is designed for developers and security teams who need to inspect the supply chain of AI agents.

How to use MCP-Scan: An MCP Security Scanner?

Obtain a Snyk API token and set it as the SNYK_TOKEN environment variable. Install uv on your system, then run uvx snyk-agent-scan@latest to auto-discover and scan all agents, MCP servers, and skills. You can also scan specific MCP configuration files or skill files by passing file paths as arguments.

Key features of MCP-Scan: An MCP Security Scanner

  • Auto-discovers MCP configurations, agent tools, and skills
  • Detects 15+ distinct security risks (e.g., prompt injection, tool poisoning)
  • Supports scanning of Claude, Cursor, Windsurf, Gemini CLI, and more
  • Interactive consent prompt before running any stdio MCP server
  • Background mode for enterprise-wide monitoring via Snyk Evo

Use cases of MCP-Scan: An MCP Security Scanner

  • Scan your local machine for vulnerable MCP servers and agent skills
  • Integrate security into CI/CD pipelines (non‑interactive mode)
  • Monitor the company‑wide agent supply chain from a central dashboard

FAQ from MCP-Scan: An MCP Security Scanner

Do I need a Snyk account to use it?

Yes. You must sign up at Snyk, obtain an API token, and set it as the SNYK_TOKEN environment variable.

What happens when scanning MCP configuration files?

Scanning an MCP config executes the commands defined in it to retrieve tool descriptions. Agent Scan prompts for user consent before starting each stdio server; for non‑interactive environments use --dangerously-run-mcp-servers.

Is my data shared with Snyk?

Yes. Skills, agent applications, tool names, and descriptions are shared with Snyk for security analysis. Agent Scan does not store or log the contents or results of MCP tool calls.

Which agents are supported?

Claude Desktop/Code, Cursor, Windsurf, VS Code, Gemini CLI, Amazon Q, and several others across macOS, Linux, and Windows.

What are the runtime requirements?

Python (via PyPI package snyk-agent-scan), the uv tool, and network access to Snyk’s API.

「開発者ツール」の他のコンテンツ