🗓️ Meeting Prep Agent

Download a full demo video by clicking here

👋 Welcome to the Tavily Meeting Prep Agent!

This repository demonstrates how to build a meeting preparation agent with real-time web access, leveraging Tavily's advanced search capabilities. This agent will connect to your Google Calendar via MCP, extract meeting information, and use Tavily search for profile research on the meeting attendees and general information on the companies you are meeting with.

The project is designed for easy customization and extension, allowing you to:

• Integrate proprietary or internal data sources
• Modify the agent architecture or swap out LLMs

---

🚀 Features

• 🌐 Real-time Web Search: Instantly fetches up-to-date information using Tavily's search API.
• 🧠 Agentic Reasoning: Combines MCP and ReAct agent flows for smarter, context-aware responses.
• 🔄 Streaming Substeps: See agentic reasoning and substeps streamed live for transparency.
• 🔗 Citations: All web search results are cited for easy verification.
• 🗓️ Google Calendar Integration: (via MCP) Access and analyze your meeting data.
• ⚡ Async FastAPI Backend: High-performance, async-ready backend for fast responses.
• 💻 Modern React Frontend: Interactive UI for dynamic user interactions.
• 📬 Email Context (via iGPT): Adds internal-only email context for meeting prep.

System Diagram

---

📂 Repository Structure

• Backend (backend/)
  • agent.py: Agentic flow (MCP + LangChain-Tavily ReAct agent)
• Frontend (ui/): React-based UI for meeting insights
• Server (app.py): FastAPI server for API endpoints and streaming

---

🛠️ Local Setup

Python version: 3.13.2 (local development)

Google Calendar MCP Setup

See google-calendar-mcp for full details.

Google Cloud Setup:

1. Go to the Google Cloud Console and create/select a project.
2. Enable the Google Calendar API.
3. Create OAuth 2.0 credentials:
   • Go to Credentials
   • Click "Create Credentials" > "OAuth client ID"
   • Choose "User data" for the type of data that the app will be accessing
   • Add your app name and contact information
   • Select "Desktop app" as the application type
4. Add your email as a test user under the OAuth Consent screen.
5. Create a file gcp-oauth.keys.json in the root of google-calendar-mcp directory.
6. Download your credentials and paste them in gcp-oauth.keys.json.

This file should look like:

{
  "installed": {
    "client_id": "<your-client-id>",
    "project_id": "<your-project-id>",
    "auth_uri": "<your-auth-uri>",
    "token_uri": "<your-token-uri>",
    "auth_provider_x509_cert_url": "<your-auth-provider>",
    "client_secret": "<your-secret>",
    "redirect_uris": ["http://localhost"]
  }
}

Install the MCP:

cd google-calendar-mcp
npm install

Set config path:

GOOGLE_CALENDAR_CONFIG=<absolute-path-to-project>/mcp-use-case/google-calendar-mcp/build/index.js

Run the notebook mcp-test.ipynb to check that your MCP setup is working before proceeding.

📬 Extending with Email Context (Optional)

By default, the agent prepares a meeting brief using public web research on attendees and the companies you are meeting with.

To ground the briefing in your real relationship with meeting attendees, enable iGPT. iGPT retrieves internal-only context from your connected email and returns meeting-relevant internal context passed into the LangGraph flow.

What changes when iGPT is available?

A typical enriched brief may include:

• What was last discussed with each attendee (internal threads)
• Commitments made by either side (and what’s still pending)
• Open items / unresolved questions

How iGPT is used in this agent

The agent sends company and attendee information (derived from your Google Calendar events) to iGPT and asks it to retrieve relevant prior internal context.

Key rule enforced in the iGPT step:

• Internal sources only (no public web information)

Setup using the iGPT Playground (recommended for demos)

You need two things:

1. IGPT_API_KEY (created in the Hub)
2. IGPT_USER (the user value you choose inside the Playground)

1) Create an API key

• Go to: https://igpt.ai/hub/apikeys/
• Create a new key (treat this as a secret - do not share it)

2) Connect your inbox (one-time per user)

You can still connect your email through the Playground, even if you don’t have an iGPT app yet.

1. Open the Playground:
   • https://igpt.ai/hub/playground/
2. Go to Connect Datasource
3. Choose a user value (this is your end-user identifier)
   • Example for a demo: demo_user
4. Submit the request - the Playground will return an OAuth link
5. Click the OAuth link and complete the login/consent
6. Indexing will start for that user

✅ Important: The user value you used in Connect Datasource is the same value you must use later when calling iGPT (and what you should place in IGPT_USER).

3) Use the same user when querying iGPT

In the Playground “Ask” screen (e.g. /v1/recall/ask) you will see a user field. That value must match the one you used when connecting the datasource.

Add to your .env

Set IGPT_USER to the exact same user value you used in the Playground:

IGPT_API_KEY=your-igpt-api-key
IGPT_USER=your-playground-user-value

Links

• API keys: iGPT Hub - API Keys
• Documentation: iGPT Docs
• Official website: iGPT

Backend Setup

1. Create and activate a virtual environment:

   python3 -m venv venv
   source venv/bin/activate  # On Windows: .\venv\Scripts\activate

2. Install dependencies:

   python3 -m pip install -r requirements.txt

3. Set environment variables:

   export TAVILY_API_KEY="your-tavily-api-key"
   export OPENAI_API_KEY="your-openai-api-key"
   export GROQ_API_KEY="your-groq-api-key"
   export GOOGLE_CALENDAR_CONFIG="<absolute-path-to-project>/mcp-use-case/google-calendar-mcp/build/index.js"

4. Run the backend server:

   python app.py

Frontend Setup

1. Navigate to the frontend directory:

   cd ui

2. Install dependencies:

   npm install

3. Start the development server:

   npm run start

.env file example:

TAVILY_API_KEY=your-tavily-api-key
OPENAI_API_KEY=your-openai-api-key
GROQ_API_KEY=your-groq-api-key
GOOGLE_CALENDAR_CONFIG=your-google-config

---

📡 API Endpoints

• POST /api/analyze-meetings: Handles streamed LangGraph execution

---

🤝 Contributing

Feel free to submit issues and enhancement requests!

---

📞 Contact

Questions, feedback, or want to build something custom? Reach out!

• Email: Dean Sacoransky

---

Powered by Tavily – The web API built for AI agents