Best for
- Security teams to monitor company-wide agent supply chain risk
- Developers to quickly assess security of their MCP servers and skills
- Evaluating third-party MCP configs in sandboxed environments
A
Agent Scan
Community
Security scanner for AI agents, MCP servers and agent skills to detect prompt injections and vulnerabilities.
GitHub source repository ↗
58FMRS · C
Agent Scan is a security scanning tool by Snyk focused on detecting vulnerabilities in AI agents and MCP servers. It supports multiple agents and provides a detailed detection coverage matrix. The tool emphasizes the risk of executing commands during MCP config scanning and recommends sandboxing. Due to experimental CLI output, it is not suitable for automated dependencies.
Agent Scan is a security scanning tool that discovers and scans agent components on your machine, including MCP servers and skills, for common threats such as prompt injections, sensitive data handling, or malware payloads hidden in natural language. It automatically discovers configurations for various agents including Claude Code/Desktop, Cursor, Gemini CLI, and Windsurf. The tool supports Scan Mode (CLI scan with report) and Background Mode (periodic scans reported to Snyk Evo). Note: scanning MCP configurations executes the commands defined in them, so it is recommended to run in a sandbox and carefully review consent prompts.
Tools
The tool list has not been reviewed yet.
Setup
- Sign up at Snyk and get an API token from https://app.snyk.io/account
- Set the token as an environment variable: export SNYK_TOKEN=your-api-token-here
- Run with uvx: uvx snyk-agent-scan@latest
Or download a standalone binary from GitHub Releases.
Fit and risk
Not for
- Users needing stable programmatic output (CLI output is experimental and may change)
- Production workflows depending on specific CLI output fields
- Scanning non-stdio MCP servers (only stdio is supported)
Required permissions
- Requires SNYK_TOKEN environment variable to call Snyk API
- Executes commands defined in MCP configurations (stdio server startup)
- Reads local agent configuration files (e.g., ~/.claude/skills)
Risks and side effects
- Scanning MCP configurations executes their commands, posing risk of malicious code execution
- Skills, app names, and tool descriptions are shared with Snyk (usage implies consent)
- CLI output is experimental and subject to change
- Large-scale API usage may result in account blocking
Troubleshooting
- Ensure SNYK_TOKEN is set correctly, otherwise scans will fail
- In non-interactive environments, use --dangerously-run-mcp-servers to bypass consent prompts
- If servers time out, adjust --server-timeout
- Check that MCP JSON config file paths are correct
Use cases
Scan all installed agent components on your machine for security vulnerabilities
Scan specific MCP configuration files (e.g., ~/.vscode/mcp.json)
Scan individual agent skill files (e.g., SKILL.md) or skill directories
Automate scanning in CI/CD pipelines (requires --dangerously-run-mcp-servers)
Supported clients
Claude DesktopFull support
Claude CodeFull support
CursorFull support
VS CodeFull support
WindsurfFull support
Gemini CLIFull support
Amazon QFull support
AmpFull support
KiroFull support
OpenCodeFull support
AntigravityFull support
CodexFull support
OpenClawFull support