🐚 MCP ShellKeeper

🐚 MCP ShellKeeper

🎯 The Problem

AI assistants like Cursor execute commands statelessly — each command runs in a fresh environment:

❌ ssh user@server                          # Hangs forever - no output until exit
❌ Can't run commands after SSH
❌ Each command starts from scratch
❌ No way to transfer files to/from servers
❌ Must re-authenticate for every operation

✨ The Solution

ShellKeeper transforms AI assistants into stateful operators with persistent sessions and file transfer capabilities.

🚀 Core Features

You: "SSH to server"
AI: ❌ Command hangs forever

You: "List files"
AI: ❌ Runs on local, not server

You: "Connect to my server"
AI: ✅ Establishes SSH session

You: "What files are there?"
AI: ✅ Lists files on server

You: "Go to the logs folder"
AI: ✅ Changes to /var/log

You: "Show me recent errors"
AI: ✅ Displays log content

You: "SSH to production server"
AI: ✅ Connected to prod-01
    Session: prod-01

You: "How much disk space left?"
AI: ✅ 45GB available
    (same session, no re-auth)

You: "Show me today's logs"
AI: ✅ Displaying logs...
    (same session continues)

You: "Restart the API service"
AI: ✅ Service restarted
    (still same session!)

You: "Upload my config to the server"

AI: ✅ Uploaded config.json
    /app/config/config.json

---

You: "Download the error logs"

AI: ✅ Downloaded app.log
    Saved to ./logs/app.log

---

You: "Get database backup"

AI: ✅ Downloaded backup.sql
    (even through SSH jumps!)

🎬 Real-World Example

Watch how a natural conversation becomes automated operations:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

👤 YOU ASK:
   "Debug the slow API on prod-server-01"

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🤖 AI EXECUTES:

   [Connecting...]
   → ssh user@prod-server-01
   ✅ Connected to prod-server-01

   [Investigating processes...]
   → ps aux | grep api
   📊 API process found: PID 1234, using 98% CPU

   [Checking logs...]
   → cd /var/log/api && tail -n 100 error.log
   🔍 ERROR: Memory leak in /api/v1/users endpoint

   [Downloading for analysis...]
   → Download: /var/log/api/error.log → ./local-debug/
   ✅ Downloaded 2.4MB (analyzed: 500+ memory errors)

   [Applying fix...]
   → Upload: ./fixed-config.yml → /app/config/
   ✅ Uploaded new rate-limiting config

   [Restarting service...]
   → systemctl restart api-service
   → systemctl status api-service
   ✅ Service restarted, CPU now at 12%

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🤖 AI RESPONDS:
   "Fixed! The API had a memory leak in the users endpoint.

   • Downloaded logs and found 500+ memory allocation errors
   • Applied rate-limiting configuration
   • Restarted service - CPU dropped from 98% to 12%
   • All operations completed in one persistent session"

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

✨ All in ONE session, NO re-authentication needed! ✨

📦 Installation

npm install -g mcp-shellkeeper

Configuration

{
  "mcpServers": {
    "shellkeeper": {
      "command": "npx",
      "args": ["-y", "mcp-shellkeeper"]
    }
  }
}

{
  "mcpServers": {
    "shellkeeper": {
      "command": "npx",
      "args": ["-y", "mcp-shellkeeper"]
    }
  }
}

{
  "cline.mcpServers": {
    "shellkeeper": {
      "command": "npx",
      "args": ["-y", "mcp-shellkeeper"]
    }
  }
}

Restart your AI assistant and you're ready! 🎉

💡 Use Cases

All through natural conversation with your AI! No scripts, no manual SSH juggling.

📖 Available Tools

The AI uses these tools automatically, but you can reference them for advanced use:

💡 Tip: The AI handles these automatically based on your natural language requests!

🔒 Security Best Practices

✅ DO:

• Use SSH key authentication (not passwords): ssh-keygen -t ed25519
• Jump through bastion hosts for production: ssh -J bastion.com user@prod
• Limit file upload destinations (avoid /etc, /root, .ssh/)
• Use read-only accounts for investigation
• Clean up sessions after tasks
• Audit all AI operations

❌ DON'T:

• Store passwords in commands or configs
• Upload untrusted files to production
• Download sensitive data without encryption
• Run destructive commands without verification
• Grant unnecessary permissions

🛠️ How It Works

Persistent Sessions:

• Uses PTY (Pseudo-Terminal) for full TTY emulation with state persistence
• Smart markers detect command completion automatically
• Exit codes captured for error detection
• Output parsed clean (no ANSI codes)

File Transfer:

• Base64 encoding through existing SSH sessions (no separate SCP/SFTP)
• Works through jump hosts without re-authentication
• Max 10MB, 5-minute timeout (completes early if faster)

🐛 Troubleshooting

// Increase timeout for long-running commands
terminal_execute({
  command: "npm install",
  timeout: 120000  // 2 minutes
})

// Check if SSH keys are set up correctly
ssh -v user@server

# Set up passwordless authentication
ssh-keygen -t ed25519
ssh-copy-id user@server

# Verify
ssh user@server "echo Success"

// Check if in SSH session first
terminal_execute({ command: "pwd" })  // Verify you're on remote server

// Ensure remote directory exists
terminal_execute({ command: "mkdir -p /app/uploads" })

// Then upload
terminal_upload({ local_path: "file.txt", remote_path: "/app/uploads/file.txt" })

// Verify remote file exists
terminal_execute({ command: "ls -lh /path/to/file" })

// Check permissions
terminal_execute({ command: "cat /path/to/file | wc -l" })

// Try download with absolute path
terminal_download({ remote_path: "/full/path/to/file", local_path: "./" })

// List all sessions
terminal_list_sessions()

// Close problematic session
terminal_close_session({ session_id: "stuck-session" })

// Create fresh session
terminal_new_session({ session_id: "new-session" })

🧪 Development

# Clone repository
git clone https://github.com/tranhuucanh/mcp-shellkeeper.git
cd mcp-shellkeeper

# Install dependencies
npm install

# Build
npm run build

# Test locally with stdio transport
node dist/index.js

# Test with MCP Inspector
npm run inspector

🤝 Contributing

Contributions welcome! Help make AI-assisted server management better.

1. Fork the repository
2. Create feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open Pull Request

📄 License

MIT License - see LICENSE file for details.

You can:

• ✅ Use commercially
• ✅ Modify
• ✅ Distribute
• ✅ Private use

🙏 Acknowledgments

• Built with Model Context Protocol SDK
• Uses node-pty for terminal emulation
• Inspired by the need for stateful command execution in AI workflows

📞 Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• MCP Community: Discord