arrismo kaggle mcp

A Model Context Protocol (MCP) server that exposes Kaggle dataset search, download, and EDA prompt generation to MCP clients such as Claude Desktop.

• Search Kaggle datasets by keyword.
• Download and unzip Kaggle datasets locally.
• Generate a starter Exploratory Data Analysis (EDA) prompt for a Kaggle dataset.
• Supports Kaggle credentials via environment variables or the standard kaggle.json file.
• Runs locally, in Docker, or through Smithery.

Searches Kaggle for datasets matching query and returns up to 10 results as JSON.

Returned fields include:

• ref
• title
• subtitle
• download_count
• last_updated
• usability_rating

Downloads and unzips a Kaggle dataset.

• dataset_ref: Kaggle dataset reference in owner/dataset-slug format, for example kaggle/titanic.
• download_path: Optional local output path. If omitted, files are saved to ./datasets/<dataset_slug>/.

Creates a prompt for generating basic Python EDA code for the provided Kaggle dataset reference. The prompt asks for data loading, missing-value checks, visualizations, and summary statistics.

• Python 3.10+
• Kaggle account and API token
• An MCP-compatible client

Create a Kaggle API token from your Kaggle account settings:

1. Go to https://www.kaggle.com/settings.
2. Select Create New API Token.
3. Download kaggle.json.

Use either environment variables or the standard Kaggle config file.

Create a .env file in the project root:

KAGGLE_USERNAME=your_kaggle_username
KAGGLE_KEY=your_kaggle_api_key

Place kaggle.json in the standard Kaggle location:

• macOS/Linux: ~/.kaggle/kaggle.json
• Windows: C:\Users\<Your User Name>\.kaggle\kaggle.json

On macOS/Linux, make sure the file is not world-readable:

chmod 600 ~/.kaggle/kaggle.json

git clone <repository-url>
cd kaggle-mcp

Create and activate a virtual environment:

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

Install dependencies with one of the following methods.

uv sync

pip install -r requirements.txt

With uv:

uv run kaggle-mcp

Or run the server module directly:

python src/server.py

The server communicates over MCP stdio and is intended to be launched by an MCP client.

Open Claude Desktop settings, then go to Developer > Edit Config and add this server to claude_desktop_config.json.

If installed in the project environment:

{
  "mcpServers": {
    "kaggle-mcp": {
      "command": "uv",
      "args": ["run", "kaggle-mcp"],
      "cwd": "/absolute/path/to/kaggle-mcp",
      "env": {
        "KAGGLE_USERNAME": "your_kaggle_username",
        "KAGGLE_KEY": "your_kaggle_api_key"
      }
    }
  }
}

If using kaggle.json, you can omit the env block.

Build the image:

docker build -t kaggle-mcp .

Run with credentials from .env:

docker run --rm -i --env-file .env kaggle-mcp

This repository includes smithery.yaml. Smithery starts the server over stdio and passes these configuration values as environment variables:

• kaggleUsername -> KAGGLE_USERNAME
• kaggleKey -> KAGGLE_KEY

1. Ask your MCP client: "Search Kaggle for heart disease datasets."
2. The client calls search_kaggle_datasets.
3. Choose a dataset reference from the results, for example user/heart-disease-dataset.
4. Ask: "Download user/heart-disease-dataset."
5. Ask: "Generate an EDA notebook prompt for user/heart-disease-dataset."

.
├── Dockerfile
├── README.md
├── pyproject.toml
├── requirements.txt
├── smithery.yaml
├── src/
│   ├── __init__.py
│   └── server.py
└── uv.lock

Downloaded datasets are saved under datasets/ by default. This directory is created at runtime when downloads are requested.