网站AI角色生成器

Booked Demo MCP Server

Model Context Protocol (MCP) server that analyzes websites and generates custom AI personas for voice agents.

What It Does

When a voice AI (like GoHighLevel Conversations) needs to analyze a prospect's website mid-call, it calls this MCP server's analyze_website tool. The server:

1. Takes a website URL as input
2. Uses Google Search + Firecrawl to gather business information
3. Generates a custom AI persona with:
   • Company name and branding
   • System instructions for the voice agent
   • Initial message
   • Brand colors
   • Source citations

Setup

1. Install Dependencies

cd mcp-server
npm install

2. Environment Variables

Create a .env file in the mcp-server directory:

# Required: Google Gemini API Key
API_KEY=your_google_gemini_api_key_here

# Optional: Firecrawl API Key (for deep web scraping)
firecrawl_key=your_firecrawl_api_key_here

3. Test Locally

npm start

The server runs via stdio (standard input/output), which is how MCP servers communicate with clients.

Connecting to GoHighLevel

In your GoHighLevel Conversations AI setup:

1. Add a new MCP server connection
2. Use command: node /path/to/mcp-server/src/index.js
3. Server will expose the analyze_website tool

Example GHL Workflow

User calls in → GHL AI collects website URL →
Calls MCP tool "analyze_website" with URL →
MCP server analyzes and returns persona →
GHL AI updates its system instructions →
Continues call as new custom persona

Deployment Options

Option 1: Railway

# Install Railway CLI
npm install -g @railway/cli

# Login and deploy
railway login
railway init
railway up

Option 2: Your Own Server

1. Copy the mcp-server directory to your server
2. Install Node.js 18+
3. Run npm install
4. Set environment variables
5. Run with npm start or via process manager (PM2, systemd)

Option 3: Docker (Coming Soon)

docker build -t booked-mcp-server .
docker run -d --env-file .env booked-mcp-server

Environment Variables Reference

Tool Schema

analyze_website

Input:

{
  "url": "example.com"
}

Output (Success):

{
  "success": true,
  "data": {
    "companyName": "Acme Corp",
    "systemInstruction": "You are [BOT_NAME]...",
    "initialMessage": "Hi! Thanks for calling...",
    "brandColor": "#2563EB",
    "sources": [
      {
        "title": "Acme Corp - About",
        "uri": "https://acme.com/about"
      }
    ]
  }
}

Output (Error):

{
  "success": false,
  "error": "Failed to analyze website: ..."
}

Troubleshooting

Server won't start

• Check Node.js version: node --version (must be 18+)
• Verify dependencies: npm install

Analysis fails

• Check Google API key is valid: echo $API_KEY
• Check Firecrawl key (if using): echo $firecrawl_key
• Check server logs for detailed error messages

GHL can't connect

• Verify the command path is correct
• Ensure Node.js is in the system PATH
• Check that environment variables are loaded

Development

Run in watch mode for auto-reload:

npm run dev

License

MIT

Update Mon Mar 23 15:40:02 PDT 2026