Kaggle MCP

🚀 Kaggle MCP (Model Context Protocol) Server

This repository contains an MCP (Model Context Protocol) server (server.py) built using the fastmcp library. It interacts with the Kaggle API, offering functions to search and download datasets, and generates prompts for EDA notebooks.

🚀 Quick Start

Project Structure

• server.py: The FastMCP server application. Defines resources, tools, and prompts to interact with Kaggle.
• .env.example: An example file for environment variables (Kaggle API credentials). Rename it to .env and fill in your details.
• requirements.txt: Lists the required Python packages.
• pyproject.toml and uv.lock: Project metadata and the uv package manager for locking dependencies.
• datasets/: The default directory for downloading Kaggle datasets.

📦 Installation

1. Clone the repository:

   git clone <repository-url>
   cd <repository-directory>
   

2. Create a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # Use `venv\Scripts\activate` on Windows
   # Or use uv: uv venv
   

3. Install dependencies:

   pip install -r requirements.txt
   

4. Set up Kaggle API credentials: Create or copy the .env file and add the following content:

   KAGGLE_USERNAME=your_kaggle_username
   KAGGLE_KEY=your_kaggle_key
   

Running the Server

1. Run in the terminal:

   python server.py
   

2. After the server starts, you can interact with it by configuring an MCP client or an AI agent.

✨ Features

Tools

• search_kaggle_datasets(query: str): Searches for Kaggle datasets. Returns a list of results related to the query.
• download_kaggle_dataset(dataset_ref: str, download_path: Optional[str] = None): Downloads the specified Kaggle dataset.
  • dataset_ref: In the format username/dataset-slug (e.g., kaggle/titanic).
  • download_path (optional): Specifies the download location. If not provided, it defaults to ./datasets/<dataset_slug>/.

Prompts

• generate_eda_notebook(dataset_ref: str): Generates a prompt message suitable for an AI model (e.g., Gemini) to create a basic exploratory data analysis (EDA) notebook for the specified Kaggle dataset. The prompt requests the generation of Python code covering data loading, missing value checks, visualization, and basic statistics.

Connecting to Claude Desktop

1. Open Claude > Settings > Developer > Edit Configuration > claude_desktop_config.json.
2. Add the following content:

   {
     "mcpServers": {
       "kaggle-mcp": {
         "command": "kaggle-mcp",
         "cwd": "<path-to-your-cloned-repo>/kaggle-mcp"
       }
     }
   }
   

3. Save the file and restart Claude Desktop.

💻 Usage Examples

An AI agent or an MCP client can interact with the server as follows:

Basic Usage

1. Agent: "Search for Kaggle datasets about 'heart disease'"
   • The server executes search_kaggle_datasets(query='heart disease')
2. Agent: "Download the dataset 'user/heart-disease-dataset'"
   • The server executes download_kaggle_dataset(dataset_ref='user/heart-disease-dataset')
3. Agent: "Generate an EDA notebook prompt for 'user/heart-disease-dataset'"
   • The server executes generate_eda_notebook(dataset_ref='user/heart-disease-dataset')
   • The server returns a structured prompt message.
4. Agent: (Sends the prompt to a code generation model) -> Receives EDA Python code.

Through the above steps, you can easily use this Kaggle MCP server for dataset search, download, and EDA notebook generation.