ci360-server-side-example

CI 360 MCP Content Server

Overview

The CI 360 MCP (Model Context Protocol) Content Server is a FastAPI-based orchestration layer for SAS Customer Intelligence 360. It enables:

• Personalized content delivery
• User interaction tracking
• Profile management across sessions

This system integrates with SAS CI360 APIs to deliver targeted marketing content and track user behavior in real time.

---

Core Capabilities

What It Does

1. Content Orchestration
   Middleware between clients and CI 360 APIs.
2. Profile Management
   Supports multiple identity types (e.g., customer_id, login_id).
3. Session Tracking
   Maintains sticky sessions via cookies.
4. Event Tracking
   Logs page loads, content views, clicks, and custom events.
5. Content Personalization
   Delivers tailored marketing content based on profiles and topics.

Key Features

• Sticky Profile Sessions: Persistent identity across sessions.
• Multi-Profile Support: Switch between personas (ryan, alice, sam, anonymous).
• Real-Time Event Tracking: Identity, data, viewable spots, and click-through events.
• Content API Integration: Fetch personalized content from CI360.
• Link Rewriting: Tracks offer clicks automatically.
• Health Monitoring: Built-in health checks and config validation.

---

Architecture

Directory Structure

app/
├── main.py                 # FastAPI application entry point
├── config/
│   ├── config.json        # CI 360 API configuration
│   └── profile.json       # User profile definitions
├── events/
│   ├── identity.py        # Identity event tracking
│   ├── data.py           # Data event tracking  
│   └── content.py        # Content API requests
├── profiles/
│   ├── profile_manager.py # Profile resolution logic
│   └── session.py        # Session management
├── helpers/
│   ├── config_loader.py  # Configuration loading
│   ├── json_loader.py    # Safe JSON loading
│   ├── logger.py         # Logging utilities
│   └── utils.py          # Utility functions
└── prompt-config/
    ├── openapi.spec       # OpenAPI specification for Claude LLM
    └── claude-haiku-4-5-system-prompt.txt # System prompt for Claude model

LLM Integration

The prompt-config/ folder enables Claude LLM integration:

• openapi.spec: Defines API endpoints for MCP server.
• claude-haiku-4-5-system-prompt.txt: Guides the LLM on:
  • Profile switching
  • Session management
  • Event tracking
  • Content personalization

---

API Endpoints

• POST /server → Main orchestration endpoint
• GET /health → Health check and config status
• GET /click → Click-through tracking and redirect

---

How It Works

Request Flow

1. Session Identification: Extract/create session ID from cookies.
2. Profile Resolution: Determine profile from request or session.
3. Identity Tracking: Send identity event if user is known.
4. Event Logging: Track page load and custom topic events.
5. Content Retrieval: Fetch personalized content from CI360.
6. Content Processing: Clean HTML and rewrite links.
7. Response Delivery: Return content with tracking metadata.

---

Profiles

Defined in profile.json:

{
  "ryan": {
    "id_type": "customer_id",
    "id_value": "ryan@example.com"
  }
}

Profiles supported:

• ryan: customer_id = ryan@example.com
• alice: login_id = alice@example.com
• sam: subject_id = abc1234567
• anonymous: _ci360_ (No identity tracking)

Switch profiles via request:

"user": "ryan"      // Switch to Ryan
"user": "_ci360_"    // Switch to anonymous

---

Event Types

• Identity Events: Link sessions to known identities.
• Load Events: Track page/session loads.
• Custom Events: Topic-specific interactions.
• Spot Viewable: When content is displayed.
• Spot Clicked: When offers are clicked.

---

Configuration

CI360 API Config (config.json)

{
  "domain": "<TENANT_DOMAIN>",
  "tenantId": "<TENANT_ID>",
  "spotId": "<SPOT_ID>",
  "apiEventKey": "<CUSTOM_EVENT_KEY>",
  "channel": "web",
  "mcpBaseUrl": "http://localhost:8000"
}

---

Tech Stack

• FastAPI (Python 3.10)
• httpx (Async HTTP client)
• uvicorn (ASGI server)
• Docker (Containerization)

---

Prerequisites

• Docker installed
  • Windows: Docker Desktop + WSL2
  • macOS/Linux: Follow https://docs.docker.com

---

Chatbot UI Setup (Example: Open WebUI)

docker run -d -p 3000:8080 \
  -v open-webui:/app/backend/data \
  --name open-webui \
  --restart always \
  ghcr.io/open-webui/open-webui:main

Access at: http://localhost:3000

---