Skip to content

MCP Servers

MCP (Model Context Protocol) servers are services that provide tools and resources to AI assistants, enabling them to interact with external systems like GitHub, databases, file systems, and APIs.

What Are MCP Servers?

An MCP server is a program that:

  • Implements the Model Context Protocol specification
  • Provides tools (functions) and resources (data) to AI assistants
  • Enables AI to interact with external systems safely
  • Can be written in any programming language

Common examples:

  • GitHub repository access
  • File system operations
  • Database queries
  • HTTP requests
  • Custom servers built by your team

Server Approval Workflow

All MCP servers must be approved before they can run. This protects your environment from unauthorized or malicious servers.

How It Works

  1. Detection - When a new MCP server is configured, CyberCage detects it automatically
  2. Blocking - Unknown servers are blocked until approved
  3. Review - Administrators review server details in the Dashboard
  4. Decision - Admins approve safe servers or block suspicious ones
  5. Enforcement - Decision applies across your entire organization

Approval States

StateDescriptionUser Impact
PendingAwaiting admin reviewCannot use server
ApprovedSafe to useServer runs normally
BlockedDeemed unsafe or unnecessaryCannot use server

Managing Servers in the Dashboard

The MCP Catalog in your Dashboard is where you manage all servers:

Server Details

Each server shows:

  • Name and description
  • Which application uses it
  • Who requested it
  • When it was first detected
  • Current approval status

Admin Actions

  • Approve - Allow the server to run
  • Block - Prevent the server from running
  • Review - Examine server details before deciding

Filtering Options

Find servers quickly by:

  • Status (Pending, Approved, Blocked)
  • Application (Claude, VS Code, etc.)
  • User who requested
  • Date detected

Server Transport Types

MCP servers can communicate using different transport protocols:

STDIO (Standard Input/Output)

  • Most common - Used by majority of MCP servers
  • Direct communication between IDE and server
  • Server runs as a local process
  • Examples: File system access, local tools

SSE (Server-Sent Events)

  • Network-based - For remote or cloud-hosted servers
  • Uses HTTP/HTTPS for communication
  • Supports real-time streaming
  • Examples: Cloud APIs, remote services

Why Transport Matters

  • STDIO servers run locally on your machine
  • SSE servers can run anywhere with network access
  • Both are fully monitored and protected by CyberCage
  • The transport type doesn't affect security - all traffic is analyzed

Auto-Approval Across Devices

Once approved, a server with the same configuration is automatically approved on all devices in your organization. This means:

  • Approve once, works everywhere
  • No repeated approvals for the same server
  • Consistent security across your team

Blocking Servers

When to Block

Consider blocking a server if:

  • It's from an untrusted source
  • It requests excessive permissions
  • Threat reports show malicious behavior
  • It's no longer needed by your team
  • Compliance policies prohibit it

What Happens When Blocked

When you block a server:

  1. All users immediately cannot use it
  2. Active instances are terminated
  3. Users see a clear error message
  4. An audit trail is created

Best Practices

For Administrators

Review Pending Servers Promptly

  • Check the MCP Catalog daily
  • Don't let legitimate work get blocked
  • Set up notifications for new servers

Document Your Decisions

  • Add notes when approving or blocking
  • Reference relevant policies
  • Help your team understand why

Regular Audits

  • Review approved servers periodically
  • Remove unused servers
  • Update policies as needed

For Developers

Use Trusted Servers

  • Prefer official MCP servers when available
  • Verify community servers before use
  • Document custom servers thoroughly

Communicate with Your Admin

  • Explain why you need a specific server
  • Provide documentation links
  • Report any suspicious behavior

Troubleshooting

Checking System Status

Use the CyberCage CLI to verify your system is working correctly:

bash
cybercage status

This command shows:

  • Daemon online/offline status
  • Hub connectivity status
  • Protected applications count
  • Monitored servers list
  • Last sync time

If you need to force a sync with the Hub:

bash
cybercage sync --force

Common Issues and Solutions

Server Stuck in Pending

If a server remains pending:

  1. Check system status: cybercage status
  2. Verify Hub connection is active
  3. Contact your organization administrator to review and approve
  4. If sync appears stale, run: cybercage sync --force

Approved Server Still Blocked

If an approved server won't run:

  1. Restart your application (configuration reload required)
  2. Force sync to get latest approvals: cybercage sync --force
  3. Check daemon status: cybercage status
  4. Wait 30 seconds for changes to propagate
  5. Contact your administrator if issue persists

Need a Server Approved Quickly

For urgent approvals:

  1. Contact your organization administrator directly
  2. Provide:
    • Server name and purpose
    • Business justification
    • Link to server documentation
  3. Administrator will review in Dashboard and approve if appropriate

Getting Support

For Individual Developers

  • First line of support: Contact your organization administrator
  • They will: Review server approvals, adjust policies, and escalate if needed
  • Provide them with: Output from cybercage status and description of the issue

For Organization Administrators

  • Dashboard Support: Access in-app support from your Dashboard
  • Email Support: contact@cybercage.io
  • Include in your request:
    • Organization name
    • Output from cybercage status
    • Screenshot of the issue from Dashboard
    • Steps to reproduce the problem

Emergency Support

For critical security incidents:

  • Security Hotline: Available for Enterprise customers
  • Email: contact@cybercage.io
  • Include: Threat details, affected users, and immediate impact

Next Steps

Built in Berlin, DE 🇩🇪