Claude Code 的渗透测试 MCP。AI 辅助安全测试，支持自动侦察、服务枚举和报告。
概览
Talon 通过 SSH MCP 将 Claude Code 连接到 Kali Linux（或任何渗透测试）虚拟机，实现 AI 辅助渗透测试，包括自动检测、结构化枚举和专业报告——全部在你的终端完成。 Talon是CarbeneAI发布的四个开源AI安全概念验证之一，另外两个是Harbinger（人工智能威胁情报）、Specter（AI辅助SIEM）和Forge（人工智能运营平台）。它们不是产品本身。它们的存在是为了让潜在客户看到CarbeneAI创始人Clint P. Garrison理解AI在安全运营中的端到端战略。 CarbeneAI是一家为风险驱动环境中CEO和董事会提供兼职CTO + CISO + AI战略咨询的机构。Clint在客户桌上拥有战略和风险席位;CarbeneAI的团队和合作伙伴网络负责执行。 使用这个仓库。分叉它。改进它。对于CarbeneAI的咨询参与：carbene。
概述（AI 翻译）
Claude Code 渗透测试 MCP
Talon 通过 SSH MCP 将 Claude Code 连接到 Kali Linux（或任何渗透测试）虚拟机，实现 AI 辅助渗透测试，包括自动检测、结构化枚举和专业报告——全部在你的终端完成。
关于CarbeneAI
Talon是CarbeneAI发布的四个开源AI安全概念验证之一，另外两个是Harbinger（人工智能威胁情报）、Specter（AI辅助SIEM）和Forge（人工智能运营平台）。它们不是产品本身。它们的存在是为了让潜在客户看到CarbeneAI创始人Clint P. Garrison理解AI在安全运营中的端到端战略。
CarbeneAI是一家为风险驱动环境中CEO和董事会提供兼职CTO + CISO + AI战略咨询的机构。Clint在客户桌上拥有战略和风险席位;CarbeneAI的团队和合作伙伴网络负责执行。
使用这个仓库。分叉它。改进它。对于CarbeneAI咨询参与：carbene.ai
它的作用
Talon 为 @anthropic-ai/mcp-server-ssh 服务器打包了完整的渗透测试方法：
- AI导向侦查 — 让Claude运行nmap、gobuster或nikto，获得结构化分析
本页由 AI 翻译，技术细节请以原文为准。
README
以下为原文 README 节选，完整内容请查看原文。
Talon
Penetration Testing MCP for Claude Code
Talon connects Claude Code to a Kali Linux (or any pentesting) VM via SSH MCP, enabling AI-assisted penetration testing with automated recon, structured enumeration, and professional reporting — all from your terminal.
About CarbeneAI
Talon is one of four open-source AI security proofs of concept published by CarbeneAI, alongside Harbinger (AI threat intel), Specter (AI-assisted SIEM), and Forge (AI operations platform). They are not the product. They exist so prospects can see CarbeneAI’s founder, Clint P. Garrison, understands the end-to-end strategy of AI in security operations.
CarbeneAI is a Fractional CTO + CISO + AI Strategy advisory practice for CEOs and Boards in risk-driven environments. Clint owns the strategy and risk seat at the client table; CarbeneAI’s team and partner network deliver execution.
Use this repo. Fork it. Improve it. For CarbeneAI advisory engagement: carbene.ai
What It Does
Talon packages the @anthropic-ai/mcp-server-ssh server with a complete penetration testing methodology:
- AI-directed recon — Ask Claude to run nmap, gobuster, or nikto and get structured analysis
- Automated 5-phase reconnaissance — Port discovery, service enumeration, web scanning, SMB, UDP
- 13-service enumeration guide — Comprehensive commands for every common service
- OSCP-style report template — Professional reporting structure for client and certification use
- Obsidian vault integration — Engagement and target note templates for organized documentation
The key insight: Claude can execute commands on your Kali VM, interpret the output, suggest next steps, and maintain a complete attack narrative — without leaving your editor.
Features
| Feature | Description |
|---|---|
| SSH MCP connection | Claude Code runs commands directly on your pentest VM |
| Automated recon | 5-phase bash script covering all initial recon steps |
| Service enumeration | 13 services: FTP, SSH, SMB, HTTP, Kerberos, LDAP, MSSQL, MySQL, RDP, WinRM, Redis, PostgreSQL, SMTP |
| Report generation | OSCP-style template with findings, evidence, and remediation structure |
| Obsidian templates | Engagement tracker and per-target notes with attack path mermaid diagrams |
| Dark workflow | Built for terminal-first, low-noise operation |
Requirements
- Claude Code installed and running
- SSH access to a Kali Linux (or other pentesting) VM
- Node.js / npx (for the MCP server)
- Target systems you have explicit authorization to test
安装
npx -y @anthropic-ai/mcp-server-ssh ssh://YOUR_USERNAME@YOUR_KALI_IP配置
{
"mcpServers": {
"kali-ssh": {
"command": "npx",
"args": [
"-y",
"@anthropic-ai/mcp-server-ssh",
"ssh://YOUR_USERNAME@YOUR_KALI_IP"
],
"description": "Kali Linux pentesting VM for security testing"
}
}
}