MCP Chatbot Setup

1. Create a virtual environment
2. Install dependencies
3. Configure your .env file
4. Run streamlit run app.py

MCP Chatbot 🧠 (Model Context Protocol Integration)

This project is a fully working chatbot powered by Azure OpenAI GPT-4o and designed to follow the principles of the MCP (Model Context Protocol), enabling seamless integration of contextual data from various sources like:

• 📬 Outlook Emails (Inbox & Sent)
• 📁 Local Files (PDFs, Excel, DOCX, etc.)
• ☁️ Google Drive (read-only content)
• 💬 Microsoft Teams (chat support in progress)

🌐 Streamlit Chat UI

Run locally:

streamlit run app.py

📚 Key Features

🚀 Project Setup

1. Clone the Repository

git clone https://github.com/<your-username>/mcp-chatbot.git
cd mcp-chatbot

2. Create Virtual Environment

conda create -n MCP_Pro python=3.10 -y
conda activate MCP_Pro

3. Install Dependencies

pip install -r requirements.txt

4. Environment Variables

Create a file named .env:

AZURE_OPENAI_ENDPOINT=https://<your-endpoint>.openai.azure.com/
DEPLOYMENT_NAME=gpt-4o-mini
API_KEY=your_azure_openai_key
OUTLOOK_CLIENT_ID=your_outlook_app_id
OUTLOOK_TENANT_ID=your_tenant_id
OUTLOOK_REDIRECT_URI=https://login.microsoftonline.com/common/oauth2/nativeclient

☁️ Google Drive Setup

• Go to Google Cloud Console
• Set up OAuth consent screen > Add scopes > Add test users
• Create OAuth Client ID (Desktop App) and download credentials.json
• Save it in the root project directory

Run:

python test_drive.py

This creates a token_drive.pkl for future Drive access.

📁 Project Structure

app.py                      # Main Streamlit UI
ask_ai.py                  # GPT-4o response builder using context
mcp_outlook.py             # Outlook email integration
mcp_google_drive.py        # Google Drive integration
local_utils.py             # Local file reading (PDF, DOCX, XLSX)
mcp_team.py                # Teams message support (coming soon)
requirements.txt           # Python dependencies
.env.template              # Sample .env file
README.md                  # This file

📌 Purpose

This project showcases how the MCP (Model Context Protocol) can be implemented in a real-world application to allow GPT models to consume dynamic, multi-source context — including emails, documents, and chat systems.

🛡️ Disclaimer

This chatbot runs locally and does not store any uploaded content. Ensure proper token and data handling before deploying to production.

👨‍💻 Built By

Created by Parth Gajmal

• GPT-4o via Azure OpenAI
• Microsoft Graph API
• Google OAuth
• MCP context pipeline architecture
• Streamlit for UI

🛣️ Roadmap

• Outlook email context integration
• Local file context extraction
• Google Drive integration
• Teams chat integration (direct + channel)
• Additional connectors (e.g. SharePoint, OneDrive)

🔮 Future Enhancements

• Add Slack, Jira, Notion
• Connect to databases (MySQL, PostgreSQL, MongoDB, etc.)
• Add user roles & access control
• RAG + Vector DB integration (FAISS, Pinecone, Weaviate, etc.)
• Integrate with SharePoint
• Support for OneDrive
• Connect to Dropbox, Box, iCloud Drive
• Support for Trello, Asana, ClickUp
• CRM integration: Salesforce, Zoho, HubSpot
• Embed inside Microsoft Teams as a bot
• Integrate with WhatsApp, Telegram, Discord
• Support CSV and JSON structured data
• Enable scheduled context refresh from connected sources
• Build admin dashboard for monitoring & analytics

Note: This is a custom implementation of the Model Context Protocol (MCP) design. It is not using any official MCP SDK or library, but follows the architectural pattern of connecting LLMs to external data sources in a standardized, context-aware way.

🧠 Are We Using the MCP Library? No, we're not using the official mcp Python library from Anthropic (or any other vendor) — because:

That library is still in very early dev or limited release.

Our chatbot uses our own MCP-style implementation, not that exact mcp module.

✅ Are We Implementing the MCP Protocol / Pattern? YES! 100% we are implementing the MCP (Model Context Protocol) conceptually and architecturally:

✔️ We’re fetching context from multiple sources like Outlook, Google Drive, Local Files, and Teams. ✔️ We format it into prompt context dynamically before calling the LLM (GPT-4o). ✔️ We use a consistent, plug-and-play method to bring context into the model — exactly what MCP aims to solve. ✔️ This fits the “open protocol to connect GPTs to tools, data, and memory” vision.

So this is a custom MCP-compatible solution — not tied to Anthropic's exact codebase, but fully aligned with the MCP philosophy.

🧾 License

MIT License