mcp-scan
Agent Scan is a security scanning tool that discovers and scans agent components on your machine, including agents, MCP servers, and skills, for prompt injections and vulnerabilities. It helps developers and security teams identify risks in their local agent configurations. It requires a Snyk account and API token.
✨ Key features
- Auto-discovers MCP configurations, agent tools, and skills
- Scans for prompt injection, tool poisoning, and toxic flows
- Detects malware payloads and untrusted content in skills
- Supports multiple agents: Claude, Cursor, Windsurf, Gemini CLI, etc.
- Interactive consent before executing MCP server commands
- Background mode for enterprise monitoring via Snyk Evo
🎯 Use cases
- Scan your machine for vulnerable MCP servers and skills
- Check a specific MCP configuration file for security issues
- Audit all Claude skills for prompt injection or malicious content
- Monitor company-wide agent supply chain in background mode
📦 Installation
🧰 Requirements: Requires a Snyk account and API token set as SNYK_TOKEN environment variable. Runs on macOS, Linux, and Windows. Install via uvx or standalone binary.
Run with uvx
Have uv installed on your system. Choose the instructions for your CLI version.
Agent Scan v0.5.x
The examples pin v0.5.17 as a concrete v0.5.x release:
uvx snyk-agent-scan@0.5.17uvx snyk-agent-scan@0.5.17 ~/.vscode/mcp.jsonuvx snyk-agent-scan@0.5.17 ~/path/to/my/SKILL.mduvx snyk-agent-scan@0.5.17 ~/.claude/skillsAgent Scan v0.6 and later
uvx snyk-agent-scan@latestuvx snyk-agent-scan@latest ~/.vscode/mcp.jsonuvx snyk-agent-scan@latest ~/path/to/my/SKILL.mduvx snyk-agent-scan@latest ~/.claude/skillsRun with a standalone binary
Download the binary for your operating system and architecture from the latest GitHub Release.
🚀 Usage
Before using, set your Snyk API token:
export SNYK_TOKEN=your-api-token-here
Then run a scan. For example, to scan the whole machine with the latest version:
uvx snyk-agent-scan@latest
To scan a specific MCP configuration file:
uvx snyk-agent-scan@latest ~/.vscode/mcp.json
⚠️ Good to know
CLI output is experimental and subject to change; scanning MCP configurations executes the commands defined in them, so run in a sandbox when scanning untrusted configs.
❓ FAQ
How do I get an API token?
Sign up at Snyk and get an API token from https://app.snyk.io/account (API Token → KEY → click to show). Set it as SNYK_TOKEN environment variable.
Does Agent Scan execute MCP server commands during scanning?
Yes, scanning an MCP configuration starts the stdio MCP servers by executing the commands defined in the config. It prompts for consent by default; use --dangerously-run-mcp-servers to bypass in trusted environments.
Which agents are supported?
Agent Scan supports Windsurf, Cursor, VS Code, Claude Desktop, Claude Code, Gemini CLI, OpenClaw, Amp, Kiro, OpenCode, Antigravity, Codex, and Amazon Q, with varying coverage by OS and scope.
Can I ignore skills during scanning?
Yes, use the --no-skills flag to ignore analysis on skills.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.