Sponsored by Deepsite.site

Claude MCP Approval Server

Created By
adstastic8 months ago
MCP server with Whatsapp integration (Twilio) to remote approve Claude Code tool use
Content

Claude MCP Approval Server

Claude MCP Approval Server Demo

A Model Context Protocol (MCP) server that provides approval prompts via WhatsApp before executing potentially sensitive commands. Built with FastMCP and Twilio.

Features

  • 🔐 Permission Prompts: Intercepts tool executions and requests approval via WhatsApp
  • 📱 WhatsApp Integration: Uses Twilio to send approval requests with quick-reply buttons
  • FastMCP: Built on FastMCP for easy integration with Claude
  • 🎯 Smart Formatting: Formats commands and reasons in a readable way
  • ⏱️ Auto-expiry: Requests expire after 5 minutes for security
  • 💾 Database Tracking: SQLite database to track approval requests

Setup Instructions

1. Install Dependencies

uv install

2. Set Up Twilio Account

  1. Sign up to Twilio at https://www.twilio.com
  2. Try WhatsApp messaging - Go to https://console.twilio.com/us1/develop/sms/try-it-out/whatsapp-learn and follow the setup guide to:
    • Join the WhatsApp Sandbox by sending a message to the Twilio number
    • Test sending/receiving messages from your device
    • Note your sandbox WhatsApp number (e.g., whatsapp:+14155238886)

3. Get Twilio Credentials

  1. Get your Account SID from the Twilio Console dashboard
  2. Generate an Auth Token at https://console.twilio.com/us1/account/keys-credentials/api-keys
    • Click "Create API Key"
    • Note the SID and Secret (this is your Auth Token)

4. Configure Environment

cp .env.example .env

Edit .env with your credentials:

TWILIO_ACCOUNT_SID=your_account_sid_here
TWILIO_AUTH_TOKEN=your_auth_token_here  
TWILIO_WHATSAPP_FROM=whatsapp:+14155238886  # Your sandbox number
APPROVAL_PHONE=+1234567890  # YOUR phone number (without whatsapp: prefix)
SERVER_PORT=8000  # Optional: Server port (defaults to 8000)

Important: Replace +1234567890 with your actual phone number that you used to join the WhatsApp sandbox. This is where approval requests will be sent.

5. Create WhatsApp Template

python setup_template.py

This automatically creates the Twilio template and updates your .env file with the template SID.

6. Expose the Server

Expose your local server using ngrok or Tailscale:

Option A: ngrok

ngrok http 8000  # Replace 8000 with your SERVER_PORT if different

Option B: Tailscale

tailscale funnel --bg 8000  # Replace 8000 with your SERVER_PORT if different

Note your public URL (e.g., https://abc123.ngrok.io or https://hostname.tail12345.ts.net)

7. Configure Twilio Webhook

  1. Go back to the Try WhatsApp page: https://console.twilio.com/us1/develop/sms/try-it-out/whatsapp-learn
  2. Click Sandbox Settings
  3. Set both webhook URLs to: https://your-public-url/twilio-webhook
    • When a message comes in: https://your-public-url/twilio-webhook
    • Status callback URL: https://your-public-url/twilio-webhook

8. Start the Approval Server

Important: Start the server BEFORE running Claude!

uv run  approval_server.py

The server should show:

  • ✅ Twilio configured
  • 🌐 Server endpoints listed
  • Webhook URL ready

9. Run Claude with Permission Prompts

claude --mcp-config mcp-servers.json \
       --mcp-debug \
       --permission-prompt-tool mcp__approval-server__permissions__approve \
       --verbose \
       --debug \
       --output-format stream-json \
       -p "your prompt here"

10. Approve/Deny from Anywhere! 🎉

When Claude tries to execute a command, you'll receive a WhatsApp message with ✅ Approve and ❌ Deny buttons. Tap to respond from anywhere in the world!

Important Caveats

⚠️ Network Access: The -p (prompt) mode has no network access ⚠️ Interactive Mode: Non-interactive mode (without -p) does not use the --permission-prompt-tool

MCP Configuration

The MCP configuration is in mcp-servers.json. Replace 8000 with your SERVER_PORT if you're using a different port.

How It Works

  1. Tool Interception: When Claude tries to execute a tool, the permissions__approve function is called first
  2. WhatsApp Notification: An approval request is sent to your WhatsApp with formatted command details
  3. User Response: You can approve or deny using the quick-reply buttons
  4. Execution: Based on your response, Claude either proceeds or stops

Message Format

Bash commands:

🔔 Approval Request

Execute command: `ls -la /etc`
*Reason:* List system configuration files

*Request ID:* abc123

⏱️ Expires in 5 minutes

Other tools:

🔔 Approval Request

*Tool:* WebFetch
  • url: https://example.com
  • prompt: Extract title

*Request ID:* abc123

⏱️ Expires in 5 minutes

Customizing the Template

To modify the WhatsApp message format or buttons:

  1. Edit template_config.json
  2. Run python setup_template.py to create a new template
  3. The script will update your .env file with the new SID

Security & Features

  • ⏱️ Auto-expiry: Requests expire after 5 minutes
  • 🔒 Phone verification: Only responses from your configured phone number are accepted
  • 📊 Audit trail: Database stores request history for tracking
  • 🔐 E2E encryption: WhatsApp messages use end-to-end encryption
  • 🌍 Global access: Approve/deny from anywhere with WhatsApp

Credits

https://github.com/mmarcen/test_permission-prompt-tool for providing a working example of using an MCP server with permission prompt tool.

Recommend Servers
TraeBuild with Free GPT-4.1 & Claude 3.7. Fully MCP-Ready.
Howtocook Mcp基于Anduin2017 / HowToCook (程序员在家做饭指南)的mcp server,帮你推荐菜谱、规划膳食,解决“今天吃什么“的世纪难题; Based on Anduin2017/HowToCook (Programmer's Guide to Cooking at Home), MCP Server helps you recommend recipes, plan meals, and solve the century old problem of "what to eat today"
Baidu Map百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
CursorThe AI Code Editor
Amap Maps高德地图官方 MCP Server
Jina AI MCP ToolsA Model Context Protocol (MCP) server that integrates with Jina AI Search Foundation APIs.
Serper MCP ServerA Serper MCP Server
Tavily Mcp
Playwright McpPlaywright MCP server
Zhipu Web SearchZhipu Web Search MCP Server is a search engine specifically designed for large models. It integrates four search engines, allowing users to flexibly compare and switch between them. Building upon the web crawling and ranking capabilities of traditional search engines, it enhances intent recognition capabilities, returning results more suitable for large model processing (such as webpage titles, URLs, summaries, site names, site icons, etc.). This helps AI applications achieve "dynamic knowledge acquisition" and "precise scenario adaptation" capabilities.
MCP AdvisorMCP Advisor & Installation - Use the right MCP server for your needs
Visual Studio Code - Open Source ("Code - OSS")Visual Studio Code
AiimagemultistyleA Model Context Protocol (MCP) server for image generation and manipulation using fal.ai's Stable Diffusion model.
DeepChatYour AI Partner on Desktop
ChatWiseThe second fastest AI chatbot™
Y GuiA web-based graphical interface for AI chat interactions with support for multiple AI models and MCP (Model Context Protocol) servers.
TimeA Model Context Protocol server that provides time and timezone conversion capabilities. This server enables LLMs to get current time information and perform timezone conversions using IANA timezone names, with automatic system timezone detection.
MiniMax MCPOfficial MiniMax Model Context Protocol (MCP) server that enables interaction with powerful Text to Speech, image generation and video generation APIs.
WindsurfThe new purpose-built IDE to harness magic
BlenderBlenderMCP connects Blender to Claude AI through the Model Context Protocol (MCP), allowing Claude to directly interact with and control Blender. This integration enables prompt assisted 3D modeling, scene creation, and manipulation.
EdgeOne Pages MCPAn MCP service designed for deploying HTML content to EdgeOne Pages and obtaining an accessible public URL.