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

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

Quick Start

1. Install the SSH MCP Server

# No global install needed — npx handles it at runtime
# Verify npx is available:
npx --version

2. Configure Your SSH Connection

Copy the example config to your Claude Code MCP config:

cp mcp-config.example.json ~/.claude/mcp.json
# Then edit with your actual SSH host and username

Or add the kali-ssh entry to your existing mcp.json:

{
  "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"
    }
  }
}

3. Set Up SSH Key Auth

Talon works best with key-based authentication (no password prompts):

# Generate a key if you don't have one
ssh-keygen -t ed25519 -C "claude-code-pentest"

# Copy to your Kali VM
ssh-copy-id YOUR_USERNAME@YOUR_KALI_IP

# Test the connection
ssh YOUR_USERNAME@YOUR_KALI_IP "uname -a"

4. Copy Recon Script to Your VM

scp scripts/recon.sh YOUR_USERNAME@YOUR_KALI_IP:~/recon.sh
ssh YOUR_USERNAME@YOUR_KALI_IP "chmod +x ~/recon.sh"

5. Start Testing

Open Claude Code and start asking:

• "Run a full recon on 10.10.10.100 using Kali"
• "Enumerate SMB shares on the target"
• "Check for web vulnerabilities on http://10.10.10.100"
• "Run linpeas and analyze the output"
• "Generate a pentest report for this engagement"

See docs/usage.md for complete workflow examples.

Repository Structure

Talon/
├── mcp-config.example.json      # Template MCP config — copy and edit
├── .env.example                 # SSH environment variables template
├── CLAUDE.md                    # Claude Code project instructions
├── scripts/
│   └── recon.sh                 # Automated 5-phase reconnaissance
├── references/
│   ├── service-enum.md          # Per-service enumeration commands (13 services)
│   └── report-template.md       # OSCP-style report template
├── templates/
│   ├── engagement.md            # Obsidian engagement tracker
│   └── target.md                # Obsidian per-target notes
└── docs/
    ├── setup-kali.md            # Kali VM setup guide
    ├── ssh-access.md            # SSH tunnel and VPN access methods
    └── usage.md                 # Usage examples and workflows

Methodology

Talon follows a structured 5-phase approach:

1. RECON  →  2. ENUM  →  3. EXPLOIT  →  4. POST-EXPLOIT  →  5. REPORT
     |             |            |               |                  |
  Passive      Services     Initial          PrivEsc           Document
  + Active     + Vulns      Access           + Loot            Findings

The scripts/recon.sh script automates Phase 1 and feeds directly into Phase 2 using the references/service-enum.md guide.

Example Prompts for Claude

Once the MCP server is connected, try these prompts:

# Initial recon
"Run the recon script on 10.10.10.50 and tell me what's interesting"

# Service deep dive
"Port 445 is open. Enumerate SMB using null session and check for vulnerabilities"

# Web testing  
"The target has a web server on port 80. Run a full web enumeration and check for common vulns"

# Post-exploitation
"I have a shell as www-data. Run linpeas and identify privilege escalation paths"

# Reporting
"Summarize all findings from this session into a professional report"

Security and Ethics

This tool is for authorized security testing only.

• Always obtain written authorization before testing any system
• Never use against systems you do not own or have explicit permission to test
• Comply with your local laws and the terms of any bug bounty programs
• Store engagement data securely and dispose of it properly after completion

Obsidian Integration

Talon includes Obsidian-compatible templates for maintaining organized pentest notes:

1. Copy templates/engagement.md to your Obsidian vault as a new note
2. Copy templates/target.md for each machine you test
3. Link targets from the engagement note for a connected knowledge graph

The templates use Obsidian callouts, mermaid diagrams, and checkbox lists for efficient note-taking during active engagements.

Related Projects

• PAI (Personal AI Infrastructure) — The open-source AI infrastructure system Talon was extracted from
• @anthropic-ai/mcp-server-ssh — The underlying SSH MCP server

Contributing

Pull requests welcome. Focus areas:

• Additional service enumeration guides
• Report template variants (PTES, bug bounty, etc.)
• Obsidian workflow improvements
• Additional automation scripts

License

MIT License. Copyright 2026 CarbeneAI.

See LICENSE for full text.

Built by CarbeneAI — Security Built In, Not Bolted On.