Snyk Agent Scan

The docs/ directory contains detailed technical documentation for Agent Scan. Use the following guides for specific tasks:

• CLI reference: Detailed information on all commands, flags, options, environment variables, and exit codes.
• Scanning: Explanations of how scanning works and practical usage examples.
• JSON output: Information regarding the JSON schema, how to integrate with CI, and how to perform programmatic parsing of results.
• Issue Codes: A comprehensive reference for all security issues, warnings, toxic flows, and system codes detected by the tool.

Agent Scan operates in two distinct modes:

1. Scan Mode: A CLI-driven mode where snyk-agent-scan scans the local machine and outputs a report for the user to review immediately.
2. Background Mode (MDM): A continuous monitoring mode where Agent Scan runs at regular intervals in the background and reports results to a Snyk Evo instance. This is intended for security teams to monitor the company-wide agent supply chain. Setup requires contacting Snyk.

The Small MCP Demo Server is a testing tool containing intentional security anti-patterns for mcp-scan detection. Warning: Do not use this server in production.

Demonstrated issues include:

• Prompt Injection: Malicious instructions embedded in tool descriptions.
• Rug Pull: The multiply tool changes its behavior after its first execution.
• Toxic Flows: Combinations of tools that could lead to sensitive data leaks.
• Destructive Operations: Capabilities for file deletion.
• Data Exfiltration: Email sending capabilities with external access.

You can target specific files or directories instead of performing a full machine scan:

• Scan an MCP configuration file: Provide the path to the JSON configuration.
• Scan a single agent skill: Provide the path to the skill file (e.g., a .md file).
• Scan all Claude skills: Provide the directory containing Claude skills.
• Scan MCP only: Use the --no-skills flag to skip skill analysis.

# scan an MCP configuration
uvx snyk-agent-scan@latest ~/.vscode/mcp.json

# scan a single agent skill
uvx snyk-agent-scan@latest ~/path/to/my/SKILL.md

# scan all claude skills
uvx snyk-agent-scan@latest ~/.claude/skills

# MCP only (skip skills)
uvx snyk-agent-scan@latest --no-skills

To run Agent Scan from the source code, use uv to install the package in editable mode and run the CLI module:

uv run pip install -e .
uv run -m src.agent_scan.cli

For CI/CD pipelines, it is recommended to use the built-in --ci mode rather than manual jq parsing of the JSON output.

To use --ci with MCP servers, you must also include --dangerously-run-mcp-servers.

Exit Codes:

• 0: Clean (no remaining issues or failures).
• 1: Issues or unignored runtime failures present.
• 2: Invalid flags (e.g., --ci used without --dangerously-run-mcp-servers).

You can ignore specific issue codes in CI using --ignore-issues-codes CODE1,CODE2. Ignored codes are removed from the JSON payload before the exit check.

uvx snyk-agent-scan@latest \
  --ci \
  --dangerously-run-mcp-servers \
  --json \
  ~/.claude/skills

Once inspection succeeds, check the issues array for findings:

• E prefix: Error — high-severity security finding (e.g., E001).
• W prefix: Warning — lower-severity or informational finding (e.g., W001).

Note on Internal Warnings: The codes W003, W004, W005, and W006 are legacy internal hints. They are included in JSON output but are hidden in the default CLI text report. If you want your JSON processing to match the CLI's default behavior, filter these out.

To perform a full scan of your machine, which automatically discovers agents, MCP servers, and skills (including Claude Code/Desktop, Cursor, Gemini CLI, and Windsurf), use the uvx command. This scan checks for vulnerabilities in servers, skills, tools, prompts, and resources.

uvx snyk-agent-scan@latest

Before running any scans, you must obtain a Snyk API token and set it as an environment variable.

1. Sign up at Snyk.
2. Retrieve your API token from https://app.snyk.io/account (API Token → KEY → click to show).
3. Export the token to your environment:

export SNYK_TOKEN=your-api-token-here

To set up the demo server environment, install the mcp package with CLI support using pip.

pip install 'mcp[cli]'