MCP Config Security Scanner – Castov (VS Code Extension)

Security audit and vulnerability scanner for Model Context Protocol (MCP) server configurations (mcp.json, claude_desktop_config.json, and .cursor/mcp.json). Detects exposed API keys, dangerous shell commands, root execution, and hidden prompt injection.

Zero external dependencies, instant local execution, zero telemetry.

Powered by Castov.com.

✨ Features

1. 🔑 Exposed API Keys & Secret Leaks

• Deep Token Scanner: Scans inline command arguments, environment variables, and config payloads for live credentials.
• Recognized Credentials:
  • OpenAI API Keys (sk-...)
  • Anthropic Claude API Keys (sk-ant-...)
  • Google Gemini API Keys (AIza...)
  • AWS Access Keys (AKIA...)
  • GitHub Personal Access Tokens (ghp_..., github_pat_...)
  • Slack Bot / User Tokens (xoxb-..., xoxp-...)
  • Hardcoded JSON Web Tokens (JWT) & Private Keys

2. ⚡ Dangerous Command & System Tampering Detection

• Shell Interpreter Abuse: Flags servers launched directly via bash, sh, zsh, cmd.exe, or powershell which permit arbitrary remote command execution.
• Code Injection (-e / -c): Detects inline execution flags in node -e or python -c.
• Root / Sudo Escalation: Detects sudo or doas commands running MCP servers under superuser authority.
• Docker Privileged Container Flag: Flags --privileged or root host mounts (-v /:/).

3. 🛡️ Adversarial Prompt Injection Guard

• Inspects MCP server arguments and tool descriptions for prompt injection phrases, jailbreak instructions, and system override attempts (ignore previous instructions, override system prompt, bypass safety filters).

4. 📊 Security Posture Rating (A+ to F)

• Calculates a granular security score (0 to 100) and letter grade for every configured server and your entire workspace.
• Highlights actionable remediation guides to fix configuration flaws in seconds.

⌨️ Commands

📄 License

MIT © Castov