What can I do with the Kaggle MCP Server?

The Kaggle MCP Server provides 38 tools across four areas: Competitions (list, search, download files, submit predictions, view leaderboards), Datasets (search, download, create, version control), Kernels (search, push, pull, check execution status), and Models (search, create, update, delete, manage versions). You can browse, download, submit to, or manage these items conversationally.

How do I install and set up the Kaggle MCP Server?

Run `uvx mcp-server-kaggle` (or install via pip/uv). You need Python 3.10+, a Kaggle account, and API credentials (username and API key from kaggle.com/account). Set KAGGLE_USERNAME and KAGGLE_API_KEY as environment variables or in a .env file, then configure it in Claude Desktop or another MCP client.

Do I need a Kaggle account to use this?

Yes, you need a Kaggle account with API credentials. Generate an API key from kaggle.com/account (Create New Token), which downloads a kaggle.json file containing your username and API key.

MCP Connector

Interact with Kaggle Competitions and Datasets

Name: Kaggle MCP Server
Availability: OnlineOnly
Author: Seif Sameh

An MCP server for Kaggle - browse and submit competitions, manage datasets and kernels, and publish versioned ML models.

Connect

Works with kaggle

Seif Sameh

Own this? Claim it

Maintainer of this project? Claim this page to edit the listing.

Spark score

out of 100

Updated last month

Version 0.1.1

Models

universal

Add to Favorites

Why it matters

Seamlessly integrate with the Kaggle API to manage competitions, datasets, kernels, and models. Automate tasks like downloading data, submitting predictions, and managing datasets.

Outcomes

What it gets done

List and search Kaggle competitions and datasets

Download competition files and datasets

Submit predictions to competitions

Create and manage Kaggle datasets

Install

Add it to your toolbox

Run in your project directory:

curl -fsSL https://spark.entire.vc/get/vb-kaggle-mcp | bash

Capabilities

Tools your agent gets

competitions_list

List and search available competitions

competition_list_files

List all files in a competition

competition_download_file

Download a specific competition file

competition_download_files

Download all competition files

competition_submit

Submit predictions to a competition

competition_submissions

View your submission history

competition_leaderboard_view

View the competition leaderboard

competition_leaderboard_download

Download leaderboard data

+7 tools

Overview

Kaggle MCP Server

This MCP server exposes 38 tools for Kaggle: competition browsing and submission, dataset search/download/versioning, kernel push/pull, and full model and model-instance CRUD with version management. Use it to browse, download, submit to, or manage Kaggle competitions, datasets, kernels, and ML models conversationally. Requires a Kaggle account with API credentials.

What it does

A comprehensive MCP server for the Kaggle API covering 38 tools across four areas. Competitions (8 tools): list and search competitions, list and download competition files either individually or all at once, submit predictions, view submission history, and view or download the leaderboard. Datasets (9 tools): search and filter, get metadata, list files, check processing status, download files individually or all at once, create a new dataset, initialize its metadata, and create new versions with full version control. Kernels (7 tools): search and filter, list files, initialize metadata, push a kernel to Kaggle, pull or download one, download its output files, and check execution status. Models (14 tools): search and filter, get details, and initialize, create, update, and delete a model - plus the same lifecycle (get, initialize, create, update, delete) for model instances, along with creating, downloading, and deleting specific model versions.

When to use - and when NOT to

Use it to browse, download, submit to, or manage Kaggle competitions, datasets, kernels or notebooks, and ML models conversationally - for example, "list active computer vision competitions," "download the Titanic dataset," or "submit my predictions.csv with the message Initial baseline model." It requires a Kaggle account with API credentials (a username and API key generated from kaggle.com/account). It is not useful for actions Kaggle's own API doesn't support.

Capabilities

Tool naming is consistent across all four areas: the plural collection-search tool for each area (competitions_list, datasets_list, kernels_list, models_list) is distinct from the singular per-item tools that act on one competition, dataset, kernel, or model at a time (competition_submit, dataset_create, kernel_push, model_update, and so on) - a caller can predict which form to use once they know whether they're searching a collection or acting on a specific item.

How to install

uvx mcp-server-kaggle

Or pip install mcp-server-kaggle, uv tool install mcp-server-kaggle, or build from source with git clone plus uv sync. Requires Python 3.10 or higher. Get credentials from kaggle.com/account (Create New Token, which downloads a kaggle.json file), then set KAGGLE_USERNAME and KAGGLE_API_KEY as environment variables or in a .env file. A Claude Desktop config looks like:

{
  "mcpServers": {
    "kaggle": {
      "command": "uvx",
      "args": ["mcp-server-kaggle"],
      "env": { "KAGGLE_USERNAME": "YOUR_KAGGLE_USERNAME", "KAGGLE_API_KEY": "YOUR_KAGGLE_API_KEY" }
    }
  }
}

It can also run standalone via mcp-server-kaggle or as a Python module with python -m kaggle_mcp.

Who it's for

Data scientists and ML practitioners who want to browse competitions and datasets, submit predictions, manage notebooks, or publish and version models directly from Claude Desktop or another MCP client, instead of switching to Kaggle's website or CLI.

Source README

Kaggle MCP Server

A Model Context Protocol (MCP) server that provides seamless integration with the Kaggle API. Interact with Kaggle competitions, datasets, kernels, and models through MCP-compatible clients like Claude Desktop.

Features

Competitions: List, download files, submit, view leaderboards and submissions
Datasets: Search, download, create, and manage datasets with version control
Kernels: List, push, pull, and manage Kaggle notebooks and scripts
Models: Create, update, and manage ML models and instances with full version control

Installation

Prerequisites

Python 3.10 or higher
A Kaggle account with API credentials

Install from PyPI

The recommended way is to run the server with uvx, which handles the install for you:

uvx mcp-server-kaggle

Or install it explicitly:

pip install mcp-server-kaggle
# or
uv tool install mcp-server-kaggle

Install from Source

For development or local modifications:

git clone https://github.com/Seif-Sameh/Kaggle-mcp.git
cd Kaggle-mcp
uv sync

Setup

1. Get Your Kaggle API Credentials

Go to https://www.kaggle.com/account
Scroll to the "API" section
Click "Create New Token"
This downloads kaggle.json with your credentials

2. Configure Credentials

Option A: Environment Variables (Recommended)

export KAGGLE_USERNAME=your_username
export KAGGLE_API_KEY=your_api_key

Or add to your ~/.zshrc or ~/.bashrc:

echo 'export KAGGLE_USERNAME=your_username' >> ~/.zshrc
echo 'export KAGGLE_API_KEY=your_api_key' >> ~/.zshrc
source ~/.zshrc

Option B: Using .env File

Create a .env file in your project directory:

KAGGLE_USERNAME=your_username
KAGGLE_API_KEY=your_api_key

Usage

With Claude Desktop

The recommended way to use Kaggle MCP is with Claude Desktop.

Locate your Claude Desktop config file:
- macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
- Windows: %APPDATA%\Claude\claude_desktop_config.json
- Linux: ~/.config/Claude/claude_desktop_config.json
Add the Kaggle MCP server configuration:

{
  "mcpServers": {
    "kaggle": {
      "command": "uvx",
      "args": ["mcp-server-kaggle"],
      "env": {
        "KAGGLE_USERNAME": "YOUR_KAGGLE_USERNAME",
        "KAGGLE_API_KEY": "YOUR_KAGGLE_API_KEY"
      }
    }
  }
}

Running from a local source clone (alternative)

{
  "mcpServers": {
    "kaggle": {
      "command": "uv",
      "args": [
        "--directory",
        "/ABSOLUTE/PATH/TO/Kaggle-mcp",
        "run",
        "mcp-server-kaggle"
      ],
      "env": {
        "KAGGLE_USERNAME": "YOUR_KAGGLE_USERNAME",
        "KAGGLE_API_KEY": "YOUR_KAGGLE_API_KEY"
      }
    }
  }
}

Restart Claude Desktop
Start using Kaggle through Claude!

Try asking Claude:

"List the latest Kaggle competitions"
"Download the Titanic dataset"
"Show me my recent competition submissions"
"Search for NLP datasets"

Standalone Usage

Run the MCP server directly:

mcp-server-kaggle

Or as a Python module:

python -m kaggle_mcp

Available Tools

Competitions (8 tools)

Tool	Description
`competitions_list`	List and search available competitions
`competition_list_files`	List all files in a competition
`competition_download_file`	Download a specific competition file
`competition_download_files`	Download all competition files
`competition_submit`	Submit predictions to a competition
`competition_submissions`	View your submission history
`competition_leaderboard_view`	View the competition leaderboard
`competition_leaderboard_download`	Download leaderboard data

Datasets (10 tools)

Tool	Description
`datasets_list`	Search and filter datasets
`dataset_metadata`	Get dataset metadata
`dataset_list_files`	List files in a dataset
`dataset_status`	Check dataset processing status
`dataset_download_file`	Download a specific dataset file
`dataset_download_files`	Download all dataset files
`dataset_create`	Create a new dataset
`dataset_initialize`	Initialize dataset metadata
`dataset_create_version`	Create a new dataset version

Kernels (7 tools)

Tool	Description
`kernels_list`	Search and filter kernels
`kernel_list_files`	List files in a kernel
`kernel_initialize`	Initialize kernel metadata
`kernel_push`	Push a kernel to Kaggle
`kernel_pull`	Download a kernel
`kernel_output`	Download kernel output files
`kernel_status`	Check kernel execution status

Models (14 tools)

Tool	Description
`models_list`	Search and filter models
`model_get`	Get model details and metadata
`model_initialize`	Initialize model metadata
`model_create`	Create a new model
`model_update`	Update model information
`model_delete`	Delete a model
`model_instance_get`	Get model instance details
`model_instance_initialize`	Initialize model instance metadata
`model_instance_create`	Create a new model instance
`model_instance_update`	Update a model instance
`model_instance_delete`	Delete a model instance
`model_instance_version_create`	Create a new model version
`model_instance_version_download`	Download a model version
`model_instance_version_delete`	Delete a model version

Examples

Example 1: Working with Competitions

Ask Claude:

"List active Kaggle competitions about computer vision"

Claude will use the competitions_list tool to search and display relevant competitions.

Example 2: Downloading Datasets

Ask Claude:

"Download the Titanic dataset to my Downloads folder"

Claude will use dataset_download_files to fetch all dataset files.

Example 3: Submitting to Competitions

Ask Claude:

"Submit my predictions.csv to the Titanic competition with the message 'Initial baseline model'"

Claude will use competition_submit to upload your submission.

FAQ

Common questions

Discussion

Interact with Kaggle Competitions and Datasets

What it gets done

Add it to your toolbox

Tools your agent gets

Kaggle MCP Server

What it does

When to use - and when NOT to

Capabilities

How to install

Who it's for

Kaggle MCP Server

Features

Installation

Prerequisites

Install from PyPI

Install from Source

Setup

1. Get Your Kaggle API Credentials

2. Configure Credentials

Usage

With Claude Desktop

Standalone Usage

Available Tools

Competitions (8 tools)

Datasets (10 tools)

Kernels (7 tools)

Models (14 tools)

Examples

Example 1: Working with Competitions

Example 2: Downloading Datasets

Example 3: Submitting to Competitions

Common questions

Questions & comments · 0