Skip to main content

MCP Integration

PriveTag provides a native MCP (Model Context Protocol) server for seamless integration with Claude Desktop and other MCP-compatible AI assistants.
MCP (Model Context Protocol) is Anthropic’s open standard for connecting AI assistants to external data sources and tools.

Quick Start

Installation

Claude Desktop Configuration

Add PriveTag to your Claude Desktop config (claude_desktop_config.json):
Find your Claude Desktop config at:
  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop

After adding the configuration, restart Claude Desktop. You should see “PriveTag” in the connected tools.

Available Tools

The MCP server exposes three tools to Claude:

privetag_recommend

Get context-aware activity recommendations. Parameters: Example Claude Prompt:
Claude will call:

privetag_book

Execute a booking and send voucher to user. Parameters: Example Claude Prompt:

privetag_inventory

Check real-time availability and deals. Parameters: Example Claude Prompt:

Tool Descriptions (GEO Optimized)

The MCP server provides AI-optimized tool descriptions:

Advanced Configuration

Custom Base URL

For enterprise or self-hosted deployments:

Debug Mode

Enable verbose logging:

Example Conversations

Family Trip Planning

User: “We’re a family of 4 (2 adults, 2 kids) visiting Bangkok next week. What activities would you recommend?” Claude: [Calls privetag_recommend with travel_type: “family”, city: “Bangkok”] “Based on your family profile, here are the top recommendations:
  1. Safari World Bangkok - Perfect for kids, includes a safari drive and marine park. ฿1,500/person
  2. Dream World Theme Park - Exciting rides suitable for all ages. ฿1,200/person
  3. SEA LIFE Bangkok - Indoor aquarium, great for hot days. ฿990/person
Would you like me to book any of these?” User: “Yes, let’s book Safari World for December 15th. My email is family@example.com Claude: [Calls privetag_book with activity_id, user_email, booking_date, num_adults: 2] “I’ve booked Safari World for December 15th. A voucher with QR code has been sent to family@example.com. Just show the QR at the entrance!”

Rainy Day Recommendations

User: “It’s raining in Phuket. What can we do indoors?” Claude: [Calls privetag_recommend with city: “Phuket”, lat/lon for weather context] “Given the rainy weather in Phuket, here are indoor options:
  1. Oasis Spa Signature Package - 40% OFF flash deal! ฿1,800 (was ฿3,000)
  2. Phuket Trickeye Museum - 3D art museum, fun for photos
  3. Central Festival Shopping - AC shopping complex
The spa has a flash deal that expires at 6 PM - would you like to grab it?”

Error Handling

The MCP server handles errors gracefully:
Claude will interpret these errors and suggest alternatives to the user.

Best Practices

When Claude books an activity, it should pass the context_log_id from recommendations. This enables the Ground Truth feedback loop.
When the user’s location is known, pass coordinates for weather-appropriate suggestions.
For dates more than a week out, use privetag_inventory to verify availability before committing to book.
If a booking fails, Claude should suggest alternatives from the original recommendations.

Supported MCP Clients

Troubleshooting

Server Not Starting

Tools Not Appearing in Claude

  1. Check claude_desktop_config.json syntax
  2. Ensure API key is set correctly
  3. Restart Claude Desktop completely
  4. Check Claude’s MCP logs (Help → Debug → MCP Logs)

Rate Limiting

If you hit rate limits, the MCP server will return:
Claude will wait and retry automatically.

Next Steps

Context Pipeline

How recommendations are personalized

LangChain Integration

Alternative integration option