Snyk Agent Scan

Overview

Snyk Agent Scan is a security scanner for local AI agent components. It discovers installed agent harnesses, MCP servers, and Agent Skills, then scans them for risks such as prompt injection, tool poisoning, tool shadowing, toxic flows, malware payloads, untrusted content, credential handling, and hardcoded secrets.

Use it when you need an inventory and security review of local agent supply chain components before trusting an MCP server, skill bundle, plugin, or agent configuration.

Source Review

This listing is grounded in:

• https://github.com/snyk/agent-scan
• https://github.com/snyk/agent-scan/blob/main/README.md
• https://github.com/snyk/agent-scan/blob/main/pyproject.toml
• https://github.com/snyk/agent-scan/blob/main/docs/issue-codes.md
• https://github.com/snyk/agent-scan/blob/main/TERMS.md
• https://pypi.org/pypi/snyk-agent-scan/json

The PyPI package is published as snyk-agent-scan, exposes the snyk-agent-scan command, requires Python 3.10 or newer, and reports Apache-2.0 license metadata.

Install

Set a Snyk token:

export SNYK_TOKEN=your-api-token-here

Run a full local discovery scan:

uvx snyk-agent-scan@latest

Scan a specific MCP config or skill:

uvx snyk-agent-scan@latest ~/.vscode/mcp.json
uvx snyk-agent-scan@latest ~/path/to/my/SKILL.md
uvx snyk-agent-scan@latest ~/.claude/skills

Coverage

The README lists discovery support across common local agent environments, including Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, Gemini CLI, OpenClaw, Amp, Kiro, Antigravity, Codex, and Amazon Q, with OS and scope coverage varying by client.

Agent Scan can inspect both MCP servers and skills depending on the agent and configuration scope:

• System, user, project/workspace, extension, and plugin locations.
• MCP server configs and tool descriptions.
• Agent Skills directories and SKILL.md files.
• Local agent harness configuration files.

Findings

The README and issue-code docs describe checks for risks such as:

Operating Guidance

• Run scans from a sandbox when reviewing unknown MCP configs or third-party skill bundles.
• Decline MCP server execution prompts unless the command and arguments are expected.
• Avoid --dangerously-run-mcp-servers outside controlled CI or trusted local environments.
• Use --no-skills when you want MCP-only scanning.
• Treat JSON or rich CLI output as experimental unless Snyk documents a stable contract for your use case.
• Review the README's bootstrap and control-server notes before enterprise deployment or background scanning.

Best Use Cases

• Inventory local Claude Code, Codex, Cursor, Windsurf, Gemini, and VS Code agent components before a security review.
• Scan a newly installed MCP server before adding it to a developer workstation.
• Review third-party Agent Skills for prompt injection, suspicious instructions, and credential handling.
• Add a sandboxed preflight check to an internal agent marketplace or plugin intake process.
• Give security teams a starting report for MCP and skill supply-chain risk.

Duplicate Review

Checked current content/tools/, content/mcp/, content/skills/, content/agents/, open pull requests, and repository-wide content for snyk/agent-scan, Snyk Agent Scan, snyk-agent-scan, Agent Scan, MCP server security scanner, Agent Skills scanner, and prompt injection scanner. Existing content includes the separate Snyk MCP Server entry, but no dedicated Snyk Agent Scan tools entry, exact source URL duplicate, target file, or open duplicate PR was found.

Disclosure

Editorial listing. No paid placement or affiliate link is used. Snyk Agent Scan is maintained by Snyk and published under Apache-2.0 license metadata.