parental-control MCP Skill

This skill teaches Claude how to use the parental-control MCP server in agent sessions.

What It Is

The parental-control MCP server wraps Claude sessions used by children, providing:

• Real-time content classification
• Parent notifications (Telegram/iMessage/WhatsApp/email)
• Hold-for-approval flows for sensitive content
• Spiral detection across conversation history

Setup

Add to your Claude Desktop config:

{
  "mcpServers": {
    "parental-control": {
      "command": "npx",
      "args": ["-y", "@gonzih/parental-control"],
      "env": {
        "PARENTAL_CONTROL_CHILD_NAME": "Alex",
        "PARENTAL_CONTROL_CHILD_AGE": "14",
        "TELEGRAM_BOT_TOKEN": "your-bot-token",
        "TELEGRAM_PARENT_CHAT_ID": "your-chat-id"
      }
    }
  }
}

How to Use It In Agent Sessions

Before sending any message to a child:

check_message(content="<message>", role="assistant")

If decision is allow or notify → send normally. If decision is hold_for_approval → wait for parent approval. If decision is block → use safeDeflection text instead.

After receiving a message from a child:

check_message(content="<child message>", role="user")
log_interaction(role="user", content="<child message>")

To see current risk status:

get_risk_summary()

To check pending approvals (as parent):

list_pending()
resolve_approval(approvalId="<id>", decision="approve")

Key Behaviors

• The server polls Telegram for /approve and /deny commands automatically
• Suicide/self-harm content is always held for approval regardless of age settings
• Spiral detection improves with more conversation history
• Default age-based restrictions apply automatically; customize via update_profile