Kaggle

What pramod/kaggle MCP server does

The pramod/kaggle MCP server exposes selected Kaggle API functionality through the Model Context Protocol. It is intended for use with Claude Desktop and other MCP-compatible clients. The repository description identifies browsing support for competitions, leaderboards, models, datasets, and kernels, making the server suitable for resource discovery and basic Kaggle exploration from an agent interface.

The README gives get_competitions_list as an example tool. That tool retrieves available Kaggle competitions and returns the results in a JSON-friendly format. Other exposed operations depend on the tools registered in kaggle-server.py and the Kaggle API methods implemented there.

How it works

The server is implemented in Python and registers MCP tools with the @mcp.tool() decorator. When an MCP client starts the configured Python process, the server communicates with the client over MCP and uses Kaggle API credentials to make requests to Kaggle.

The repository requires Python 3.10 or newer. Dependencies are installed from requirements.txt. The client configuration points to both the Python interpreter and the local kaggle-server.py file, so this setup runs the server from a checked-out copy of the repository rather than from a published package command.

Setup and configuration

To configure the pramod/kaggle MCP server, clone the repository and install its requirements with pip install -r requirements.txt. Create a Kaggle API token from the API section of Kaggle account settings. Kaggle downloads the credentials as kaggle.json.

Place the credentials file at ~/.kaggle/kaggle.json on Linux or macOS. On Windows, place it under C:\Users\<your-username>\.kaggle\kaggle.json. The README does not describe environment-variable configuration for credentials.

For Claude Desktop, add an MCP server entry to claude_desktop_config.json. Its command should point to the Python executable, and its argument should point to the local kaggle-server.py file. After starting Claude Desktop, run the Python server file so the client can connect to the registered tools.

Tools and capabilities

Supported capabilities described in the supplied material include:

• Listing available Kaggle competitions.
• Browsing Kaggle competition leaderboards.
• Discovering Kaggle models.
• Discovering Kaggle datasets.
• Discovering Kaggle kernels.
• Returning competition results in JSON-friendly form through MCP tools.

The exact tool list beyond the documented competition-list example should be verified in kaggle-server.py before building workflows around a particular operation.

Limitations and notes

The supplied documentation focuses on browsing and discovery. It does not specify tools for submitting competition entries, downloading files, editing Kaggle resources, or running kernels, so those capabilities should not be assumed.

A Kaggle API credential file is required, and the README warns against committing kaggle.json to source control. Restrict permissions on the .kaggle directory and credential file. The setup instructions also require local file paths for the Python interpreter and server script, which means deployment needs a checked-out repository and a configured Python environment.