MCP: Cornell Resume

A Model Context Protocol (MCP) server that automatically generates Cornell-style study notes and summaries from the conversational context, with RAG active recall question generation and Notion integration.

Features

The server processes text from the conversational context, generates contextual summaries, creates active recall questions (context-aware), and automatically saves everything to your Notion.

• Real-time Cornell-style note generation from the client chat conversation history
• Context-aware active recall question generation using vector similarity
• Semantic search integration with Pinecone for relevant note retrieval
• Automatic Notion database synchronization with proper block formatting
• OpenAI-powered text processing and question generation

Setup Guide

Requirements

• Python 3.13+
• uv
• OpenAI account
• Pinecone account

Installation

1. Clone the repository

git clone
cd mcp-cornell-resume

2. Install dependencies

# Create virtual environment
uv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install required packages
uv pip sync docs/requirements.txt

3. Environment Configuration - Create a .env file in the project root:

OPENAI_API_KEY=your_openai_api_key
PINECONE_API_KEY=your_pinecone_api_key
PINECONE_INDEX_NAME=cornell-notes
NOTION_API_KEY=your_notion_api_key
NOTION_DATABASE_ID=your_database_id

4. Set up Notion

• Create a Notion Integration at the notion developers page (Set appropriate capabilities at minimum, select "Read content" and "Update content").
• Get your API Key
• Share Pages with your Integration.
• Get the database key of the page: You will find the database in in the link of the page.

5. Configure Pinecone Index - Create a Pinecone index with:

• Dimension: 1536 (for OpenAI embeddings)
• Index name matching your .env configuration

Usage

Using with Claude Client Desktop or other MCP-compatible applications

Add this to your MCP configuration JSON file:

code ~/Library/Application\ Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "resume_to_notion": {
      "command": "/Users/<USER></USER>/.local/bin/uv",
      "args": [
        "--directory",
        "/<PATH_TO_PROJECT/mcp-cornell-resume",
        "run",
        "main.py"
      ]
    }
  }
}

Available Tools

save_resume_to_notion

Summarize the full ongoing chat conversation in send it as 'text'.

Parameters:

• text: text string a resume of the client chat window.

Return:

• notion_page_id

How It Works

Process Flow

1. Input Processing: The MCP client sends chat conversation text to the save_resume_to_notion tool
2. Embedding Generation: OpenAI creates vector embeddings from the input text content
3. Context Retrieval: Pinecone searches for semantically similar existing notes using the embeddings
4. Cornell Summary Generation: OpenAI generates a structured Cornell-style summary using:
   • Original conversation content
   • Related notes from Pinecone for context awareness
5. Vector Storage: The new note is stored in Pinecone with its embeddings for future context retrieval
6. Notion Integration: The formatted Cornell summary is saved to your Notion database
7. Response: Returns the Notion page ID to the client

Visual Flow Diagram

Development

You can use the Model Context Protocol inspector to try out the server:

fastmcp dev main.py

Limitations and Future Improvements

• LLM Context Window Limitations: The MCP client has a finite context window (typically 8k-32k tokens depending on the model).
• If a chat session is too long, summarizing only the available context might lose important information.
• Latency in Multi-Integration Workflows: Each integration adds latency and increases the risk of slower processing.
• Notion and Pinecone Sync Complexity.
• Create a tag RAG feature for the notes.

Security Considerations

• API Key Management: Store all API keys securely in .env
• Input Validation: All inputs are sanitized and validated
• Error Handling: Sensitive information is never exposed in errors
• Access Control: Notion integration respects workspace permissions
• Data Privacy: No chat content is permanently stored without consent

License

MIT License