Kaggle

You can run a Model Context Protocol (MCP) server that interacts with Kaggle to search for datasets, download them, and generate prompts for Exploratory Data Analysis notebooks. This server exposes tools you can call from an MCP client to perform these tasks in a streamlined, programmatic way.

How to use

Use an MCP client to talk to the Kaggle MCP Server and perform three core actions: search Kaggle datasets, download a dataset, and generate an EDA notebook prompt. You can chain these actions to quickly locate a dataset, fetch it to a local directory, and prepare code prompts for analysis.

How to install

Prerequisites you need before starting: Python, a virtual environment tool, and a working network connection.

# Choose one of the two common setup paths
# Path A: Using Python and virtualenv
python -m venv venv
source venv/bin/activate  # On Windows use `venv\Scripts\activate`
pip install -r requirements.txt

# Path B: Using uv (alternative dependency manager)
uv sync

# Ensure you have Kaggle API credentials before running the server (see the Credentials section below) and that your environment can access Kaggle datasets.

Configuration and credentials

The server relies on Kaggle API credentials to access datasets. Use one of the following approaches to provide them.

# Method 1: Environment variables (recommended)
KAGGLE_USERNAME=your_kaggle_username
KAGGLE_KEY=your_kaggle_api_key

# Method 2: kaggle.json file
# Place kaggle.json in the default location: ~/.kaggle/kaggle.json (Linux/macOS) or C:\Users\<User>\.kaggle\kaggle.json (Windows)

Running the server

Start the Kaggle MCP Server using the provided run command. Ensure your virtual environment is active before launching the server.

uv run kaggle-mcp

Notes and security

Keep your Kaggle API credentials secure. Do not share your Kaggle key. If you are deploying in a shared or CI environment, consider using a secure secret store for credentials and inject them as environment variables at runtime.

Available tools

search_kaggle_datasets

Search Kaggle datasets using a query string and return the top results with details such as reference, title, download count, and last updated date.

download_kaggle_dataset

Download and unzip files for a specific Kaggle dataset given its reference in the format username/dataset-slug and store them to a local path if provided.

generate_eda_notebook

Create a structured prompt for an AI model to generate a basic Exploratory Data Analysis notebook for the specified dataset, including data loading, missing value checks, visualizations, and statistics.