mcp-servers/mcp-gateway/DEPLOYMENT_CHECKLIST.md

4.6 KiB

MCP Gateway + Open-UI Deployment Checklist

Completed Setup

Files Created/Modified:

  • gateway-proxy/openai_routes.py - NEW - OpenAI API handlers
  • openai_adapter.py - NEW - Adapter class (optional)
  • gateway-proxy/gateway_proxy.py - MODIFIED - Added OpenAI routes integration
  • OPENAI_INTEGRATION.md - NEW - Technical integration guide
  • SETUP_OPEN_UI.md - NEW - Quick start guide
  • DEPLOYMENT_CHECKLIST.md - NEW - This file

What Was Added:

# In gateway_proxy.py:
from openai_routes import chat_completions, list_models

# Routes added:
Route("/v1/models", openai_models, methods=["GET"]),
Route("/v1/chat/completions", openai_completions, methods=["POST"]),

🚀 Deployment Steps

On Your Server (wherever docker-compose is running):

  1. Navigate to gateway directory:

    cd /path/to/MCP\ Servers/mcp-gateway/
    
  2. Verify the files are in place:

    ls -la gateway-proxy/openai_routes.py
    ls -la openai_adapter.py
    
  3. Restart the gateway:

    docker compose restart mcp-gateway
    
  4. Verify startup was successful:

    docker compose logs mcp-gateway | grep -i "openai\|v1"
    

    Expected output:

    mcp-gateway  | 2026-03-30 09:15:42 - mcp-gateway - INFO - OpenAI-compatible endpoints registered at /v1/*
    
  5. Test the endpoint:

    curl -H "Authorization: Bearer rIoRseJSR9rqwl5bp4CEOyOUKdymaoJnAsT-y752BDi2qD9jJYbgd8sWkiQIGMtz" \
      https://mcp.wilddragon.net/v1/models
    

📋 Open-UI Configuration

After restart, configure Open-UI:

Location: Admin Panel → Settings → Connections

  1. Scroll to "Direct Connections"
  2. Toggle ON (if not already enabled)
  3. Add New Connection
  4. Fill in:
    API Base URL:  https://mcp.wilddragon.net/v1
    API Key:       rIoRseJSR9rqwl5bp4CEOyOUKdymaoJnAsT-y752BDi2qD9jJYbgd8sWkiQIGMtz
    
  5. Save

In Chat:

  • Model selection: mcp-gateway
  • Start chatting!

🔒 Security Notes

Bearer Token

  • Current token: rIoRseJSR9rqwl5bp4CEOyOUKdymaoJnAsT-y752BDi2qD9jJYbgd8sWkiQIGMtz
  • Keep secret: This token grants access to all MCP tools
  • Share carefully: Only give to trusted Open-UI instances
  • Rotate if needed: Generate a new token with the gateway's token generation

HTTPS Only

  • Ensure https://mcp.wilddragon.net has a valid SSL certificate
  • Never use http:// in production

📊 What You Get

In Claude.ai:

  • Full MCP protocol support (existing)
  • All tools available via OAuth

In Open-UI:

  • OpenAI-compatible API (/v1/*)
  • Tool listing and discovery
  • Bearer token authentication
  • Streaming responses support

Services Connected:

  • ERPNext (ERP data)
  • Wave Finance (accounting)
  • TrueNAS (storage)
  • Home Assistant (automation)

🧪 Testing

Quick Test - List Models:

curl -H "Authorization: Bearer YOUR_TOKEN" \
  https://mcp.wilddragon.net/v1/models

Full Test - Chat Request:

curl -X POST \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mcp-gateway",
    "messages": [
      {"role": "user", "content": "What tools are available?"}
    ]
  }' \
  https://mcp.wilddragon.net/v1/chat/completions

⚠️ If Something Goes Wrong

Gateway won't start:

docker compose logs mcp-gateway
# Look for import errors or Python syntax errors

OpenAI endpoints not showing:

# Check if openai_routes.py was properly created
python3 -m py_compile gateway-proxy/openai_routes.py

Can't connect from Open-UI:

# Verify gateway is running
docker compose ps mcp-gateway

# Test connectivity
curl -k https://mcp.wilddragon.net/v1/models

📞 Support

This Setup Provides:

  1. MCP protocol - For Claude.ai (existing, working)
  2. OpenAI API - For Open-UI (newly added)
  3. Bearer tokens - For authentication (both)
  4. Multiple backends - ERPNext, Wave, TrueNAS, Home Assistant

Files to Reference:

  • SETUP_OPEN_UI.md - Step-by-step guide
  • OPENAI_INTEGRATION.md - Technical details
  • gateway-proxy/openai_routes.py - Implementation
  • docker-compose.yml - Service configuration

Next Steps (Optional)

  1. Tool Execution: Implement actual tool calls (not just listing)
  2. Custom Prompts: Add system prompts to guide tool usage
  3. Logging: Track all API calls for debugging
  4. Rate Limiting: Protect against abuse
  5. Caching: Cache tool responses for better performance

You're all set! 🎉 Your MCP Gateway now works with both Claude.ai and Open-UI.