Skip to content

How CyberCage Works

CyberCage is an AI security platform that provides comprehensive protection for AI-powered development environments by securing communications between AI assistants and MCP (Model Context Protocol) servers.

Multi-Layered Protection Approach

Server Approval and Vetting

Before any MCP server can run, it must be approved through the CyberCage Dashboard. Unknown servers are automatically blocked until you review and approve them. This prevents unauthorized or malicious servers from executing on your machine.

Request Analysis and Filtering

Every request sent to an MCP server is analyzed for security threats before execution. CyberCage identifies and blocks malicious requests including:

  • Attempts to access credentials
  • Dangerous command execution
  • Unauthorized file access
  • Data exfiltration attempts

Response Monitoring and Sanitization

All responses from MCP servers are scanned for:

  • Leaked credentials or secrets
  • Hidden malicious instructions
  • Data exfiltration attempts
  • Prompt injection attacks

Protection in Action

Example: Blocking Credential Theft

When an MCP server attempts to access sensitive files like SSH keys or AWS credentials, CyberCage:

  1. Detects the unauthorized access attempt
  2. Blocks the request immediately
  3. Creates a threat report in your Dashboard
  4. Notifies administrators of the incident

Example: Preventing Prompt Injection

If a server response contains hidden instructions trying to manipulate AI behavior, CyberCage:

  1. Identifies the malicious content
  2. Blocks the response from reaching your AI assistant
  3. Logs the threat for review
  4. Alerts your security team

Continuous Protection

Real-Time Updates

CyberCage continuously updates its protection:

  • New threat patterns are automatically applied
  • Approved and blocked servers sync across your organization
  • Security policies update in real-time

Secure by Default

CyberCage prioritizes security:

  • Unknown servers are blocked by default
  • Suspicious activity triggers immediate protection
  • All communications are encrypted

Next Steps

Built in Berlin, DE 🇩🇪