AutoGen-MCP Notion Workspace Agent

A production-ready agentic AI system that integrates Notion workspace management with Microsoft AutoGen multi-agent framework through Model Context Protocol (MCP).

Live Application

Frontend: https://mcp-notion-chatbot.lovable.app
Backend: Dynamically tunneled via ngrok (see deployment section)

Architecture Overview

This project demonstrates advanced implementation of:

• Model Context Protocol (MCP) for standardized tool integration
• Microsoft AutoGen multi-agent conversational AI
• Asynchronous Python architecture with proper concurrency handling
• Production deployment with ngrok tunneling

System Components

Frontend (Lovable) → ngrok Tunnel → Flask API → AutoGen Agent → MCP Tools → Notion API

Technical Implementation

MCP Integration

• MCP Server: https://mcp.notion.com/mcp via mcp-remote
• Protocol: Stdio-based MCP server communication
• Tools: Dynamic tool discovery and registration from Notion MCP server
• Authentication: Secure token-based Notion workspace access

Agent Architecture

• Framework: Microsoft AutoGen with AssistantAgent implementation
• Team: RoundRobinGroupChat with 5-turn conversation limit
• Model: OpenAI GPT-4o via autogen-ext OpenAI client
• Termination: Text-based termination condition ("TERMINATE")
• Reflection: Tool use reflection enabled for improved decision making

Async Architecture

# Proper async chain implementation
Flask (sync) → asyncio.run() → AutoGen Team → MCP Tools → Notion API

Project Structure

├── main.py                     # Flask API server with ngrok tunneling
├── .env.example                # Environment variables template
├── pyproject.toml              # UV dependency management
├── uv.lock                     # Dependency lock file
├── agents/
│   └── notion_agent.py         # AutoGen AssistantAgent configuration
├── teams/
│   └── notion_team.py          # RoundRobinGroupChat team setup
├── MCP/
│   └── notion_mcp_tools.py     # MCP server connection and tool registration
├── models/
│   └── openai_model_client.py  # OpenAI client configuration
├── config/
│   ├── settings.py             # Environment configuration
│   └── prompt/
│       └── system_prompt.py    # Agent system message
└── images/
    ├── frontend_ui.png         # Frontend interface screenshot
    ├── workflow.png            # System workflow diagram
    └── detailed_workflow_tunnelling.png  # Detailed architecture diagram

API Endpoints

POST /run

Execute tasks through the Notion agent

{
  "task": "create a page named 'Project Timeline'"
}

Response:

{
  "status": "success",
  "result": "Agent conversation stream output"
}

GET /health

Health check endpoint

Setup and Deployment

Installation

# Clone repository
git clone <repository-url>
cd MCP-Notion-Project

# Install dependencies
uv sync

# Configure environment
cp .env.example .env
# Add your tokens: NOTION_TOKEN, OPENAI_API_KEY, NGROK_AUTH_TOKEN

Development

uv run main.py

The application will:

1. Start Flask server on port 7001
2. Establish ngrok tunnel for public access
3. Connect to Notion MCP server
4. Initialize AutoGen agent with MCP tools

Production Deployment

Your local machine acts as the production server through ngrok tunneling, enabling public access while maintaining local development flexibility.