Tengu is an open-source MCP server that exposes 80 pentesting tools from Kali Linux (Nmap, Metasploit, Nuclei, SQLMap, ZAP, and more) to AI assistants like Claude. Every tool call goes through input sanitization, target allowlisting, rate limiting, and audit logging — no shell=True, ever.

  It implements all three MCP primitives: 80 tools, 20 resources (OWASP Top 10, MITRE ATT&CK, CVE databases, default credentials, checklists), and 35
   prompts — guided workflows for full pentests, vulnerability assessments, and professional report generation. Findings are automatically correlated
   with CVEs for context and remediation guidance.

  Works as a copilot (Claude Code / Claude Desktop) or as an autonomous agent with human-in-the-loop for destructive actions. Single `docker compose
  up` to get started. Ships in 3 tiers: core, lab (with intentionally vulnerable targets), and full stack.

  74 dedicated command injection tests. MIT licensed.

  Would love feedback on the security model and MCP patterns.