MCPScan by yyyutakaaa

Version updated for https://github.com/yyyutakaaa/mcpscan to version v0.1.0.

  • This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

mcpscan is an automated security tool designed to detect common security issues in machine learning models and related configurations, such as injection attacks, dangerous code execution, misconfigured servers, exfiltration vulnerabilities, supply chain risks, and secrets exposure. It analyzes Python scripts, configs, and skill folders to provide detailed reports on potential security weaknesses, helping developers prevent threats before they reach their models.

What’s Changed

MCPScan v0.1.0

Shipped the first release. MCPScan scans MCP servers and AI agent skills for the stuff that tends to get missed in review.

Detects

  • Prompt-injection and tool-poisoning patterns
  • Hidden Unicode and bidirectional control characters
  • Dangerous Python calls like eval, exec, subprocess, and os.system
  • Unsafe file access and path traversal risks
  • Hardcoded external exfiltration endpoints
  • Embedded credentials and leaked secrets
  • MCP servers running wide open (overly permissive flags)
  • Unpinned installs and remote shell scripts piped into sh

Features

  • Terminal, JSON, and SARIF output
  • Adjustable severity threshold
  • GitHub Code Scanning integration
  • Custom YAML rules for stuff specific to your stack
  • Tested on Python 3.11, 3.12, and 3.13 This is static analysis, not proof. Findings are patterns worth a second look, not confirmed exploits — a developer or security engineer should still review them before acting.