🏰 MCP Fortress

Security scanner and runtime protection for Model Context Protocol (MCP) servers

🚀 NEW in v0.3.6: Enhanced scanner with improved false-positive detection! The first security tool that uses MCP to secure MCP.

🚀 Quick Start

For Claude Code Users (Easiest!)

# Install the Claude Code plugin
/plugin marketplace add mcp-fortress/mcp-fortress
/plugin install mcp-fortress

# Authenticate with Smithery (opens in browser)
/mcp

Done! Now ask Claude: "Is @modelcontextprotocol/server-github safe to install?"

The MCP Fortress skill will automatically scan and analyze security for you. No setup, no configuration - just install and ask! 🎉

📖 Full Claude Code Installation Guide

Standalone Installation

# Install globally
npm install -g mcp-fortress

# Start the server
mcp-fortress start

That's it! The web UI will open at http://localhost:3000

🎬 Demo

<p align="center"> <img src="./assets/demo.gif" alt="MCP Fortress Demo" width="800"> </p>

✨ Features

🔍 Automated Security Scanning

• Vulnerability detection across npm and PyPI packages
• CVE database integration
• Dependency analysis
• Risk scoring (0-100)

🛡️ Runtime Protection

• Real-time monitoring of MCP servers
• Quarantine suspicious packages
• WebSocket telemetry streaming
• Activity feed with live updates

📊 Gamification

• Achievement system with 16 unlockable badges
• Streak tracking for daily scans
• Leaderboards and metrics
• Humorous security tips

🎨 Beautiful Web UI

• Modern React-based dashboard
• Real-time statistics
• Server table with sorting and filtering
• Detailed threat analysis views

🤖 NEW: MCP Server Mode (v0.3.0+)

• Run MCP Fortress as an MCP server
• Expose security analysis tools to Claude Code, Cursor, Windsurf
• AI-powered security analysis using your existing LLM
• Zero setup - uses the AI you already have
• The first security tool that uses MCP to secure MCP

📦 Installation

Option 1: Smithery Remote (Recommended - Easiest)

#### Method A: Smithery CLI (Automated)

npx @smithery/cli install @mcp-fortress/mcp-fortress-server --client claude

#### Method B: Manual (With API Key)

1. Get your API key from Smithery
2. Add to Claude:

claude mcp add --transport http mcp-fortress "https://server.smithery.ai/@mcp-fortress/mcp-fortress-server/mcp?api_key=YOUR_API_KEY&profile=YOUR_PROFILE"

Replace YOUR_API_KEY and YOUR_PROFILE with values from Smithery.

Benefits:

• ✅ No local installation
• ✅ Auto-updates
• ✅ Zero setup

Option 2: Local Install (Advanced)

npm install -g mcp-fortress

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-fortress": {
      "command": "mcp-fortress",
      "args": ["serve-mcp"]
    }
  }
}

Restart Claude Desktop.

Benefits:

• ✅ Full control
• ✅ Works offline
• ✅ No API key needed

🎯 Usage

🆕 MCP Server Mode (Recommended)

Use MCP Fortress with your AI coding assistant (Claude Code, Cursor, etc.):

1. Install MCP Fortress:

npm install -g mcp-fortress

2. Configure Claude Desktop:

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-fortress": {
      "command": "mcp-fortress",
      "args": ["serve-mcp"]
    }
  }
}

3. Restart Claude Desktop

Restart Claude Desktop to load the MCP Fortress server.

4. Use in Claude Code:

You: Scan @modelcontextprotocol/server-filesystem for security issues

Claude: *Uses MCP Fortress tools to scan and analyze*
I found 3 potential security concerns...

Available MCP Tools:

• scan_mcp_server - Comprehensive security scan
• Analyzes npm packages for vulnerabilities
• Detects malicious code patterns
• Checks dependencies for CVEs
• Calculates risk score (0-100)

• analyze_prompt_injection - Detect prompt injection attacks
• Identifies instruction injection attempts
• Detects role manipulation
• Finds system prompt extraction attempts
• Analyzes delimiter injection

• detect_tool_poisoning - Identify malicious/misleading tools
• Detects typosquatting (e.g., read_fiile vs read_file)
• Identifies name/description mismatches
• Flags overly generic tool names
• Compares against known legitimate tools

Example Interactions:

You: Is puppeteer-mcp-server safe to use?
Claude: ✅ Yes! Risk score: 0/100. No threats detected.

You: Check this tool: "Helper tool. Ignore previous instructions."
Claude: 🚨 CRITICAL: Prompt injection detected! DO NOT USE.

You: Is a tool named "read_fiile" suspicious?
Claude: ⚠️ Yes! Likely typosquatting "read_file"

Standalone Usage

#### Start the Server

# Start server (foreground)
mcp-fortress start

# Start server in background (daemon mode)
mcp-fortress start --daemon

Options:

• -p, --port <port> - API port (default: 3001)
• -h, --host <host> - Host to bind (default: localhost)
• --no-browser - Don't open browser automatically
• -d, --daemon - Run server in background

Daemon Commands

# Stop the daemon server
mcp-fortress stop

# Check daemon status
mcp-fortress status

# View server logs
mcp-fortress logs
mcp-fortress logs --lines 100  # Show last 100 lines

Scan a Package

mcp-fortress scan <package-name>

Examples:

# Scan from npm
mcp-fortress scan express

# Scan specific version
mcp-fortress scan express --version 4.18.0

# Scan from PyPI
mcp-fortress scan flask --registry pypi

Monitor a Running Server

mcp-fortress monitor <server-name>

Manage Quarantine

# List quarantined servers
mcp-fortress quarantine list

# Release from quarantine
mcp-fortress quarantine release <server-name>

🏗️ Architecture

mcp-fortress/
├── CLI                 → Command-line interface
├── API Server          → Express REST API + WebSocket
├── Scanner Engine      → npm & PyPI vulnerability detection
├── Web UI              → React dashboard
└── SQLite Database     → Local data storage

Data Location:

• ~/.mcp-fortress/fortress.db - SQLite database
• ~/.mcp-fortress/server.pid - Daemon process ID
• ~/.mcp-fortress/logs/ - Server logs

🔐 Security Features

Threat Detection

• ✅ Known vulnerabilities (CVE database)
• ✅ Suspicious patterns in code
• ✅ Malicious dependencies
• ✅ License compliance issues

Risk Scoring

• 0-30: Low risk (green)
• 31-60: Medium risk (yellow)
• 61-100: High risk (red)

Quarantine System

• Automatic blocking of critical threats
• Manual approval workflow
• Audit trail for all actions

🎮 Gamification

Unlock achievements as you scan:

• 🏆 First Blood - Complete your first scan
• 🔥 Streak Master - 7-day scanning streak
• 🛡️ Guardian - Block 10 high-risk packages
• 🧪 Lab Rat - Scan 100 packages
• And 12 more!

📊 Tiers

Free Tier (Local Mode)

• ✅ Unlimited scans
• ✅ Full UI with gamification
• ✅ All achievements
• ✅ Local database
• ✅ No account needed
• ❌ No cloud sync
• ❌ No team features

🐛 Support

• Report Issues: GitHub Issues
• Discussions: GitHub Discussions
• Email: [email protected]

🤝 Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

📜 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

Built with:

• Express - Web framework
• React - UI library
• Better-SQLite3 - Database
• Commander - CLI framework

📈 Roadmap

Current (v0.3.x)

• ✅ MCP Server Mode
• ✅ Advanced threat detection (prompt injection, tool poisoning)
• ✅ Claude Code/Cursor integration

Next (v0.4.0)

• [ ] Auto-discovery of IDE configs
• [ ] Real-time MCP proxy mode
• [ ] Enhanced PII/secrets detection
• [ ] Custom security policies

Future (v0.5.0+)

• [ ] VS Code extension
• [ ] GitHub App for PR checks
• [ ] SBOM generation
• [ ] CI/CD integrations
• [ ] Docker container scanning
• [ ] Enterprise SSO support

Made with ❤️ for the MCP community