kaggle

Kaggle-MCP enables Claude AI to interact with Kaggle via the Model Context Protocol (MCP). Once running, Claude can authenticate with Kaggle and use tools to browse competitions, search and download datasets, explore kernels (notebooks), and access pre-trained models available on Kaggle. The available capabilities include authentication, competitions, datasets, kernels, and models, allowing you to perform tasks such as listing active competitions, downloading datasets, and analyzing kernels directly from Claude’s interface. To get started, ensure your Kaggle API credentials are configured so Claude can authenticate and access Kaggle resources. You can authenticate by providing your Kaggle username and API key when prompted or via the built-in authenticate() tool in Claude. The MCP server exposes a set of commands and tools documented in the project resources, which guide you through common tasks like listing competitions, finding datasets about a topic, or locating kernels related to a subject.

Prerequisites:\n- Python 3.8 or newer\n- pip (often comes with Python)\n- MCP Python SDK (as required by the Kaggle-MCP server)\n- Optional: uv (for accelerated Python environments)\n\n1) Install the Kaggle-MCP package:\n\n- macOS / Linux (via installer script):\n curl -LsSf https://raw.githubusercontent.com/54yyyu/kaggle-mcp/main/install.sh | sh\n\n- Windows (via installer):\n powershell -c "Invoke-WebRequest -Uri https://raw.githubusercontent.com/54yyyu/kaggle-mcp/main/install.ps1 -OutFile install.ps1; .\install.ps1"\n\n2) Manual installation (alternative):\n\n- Using pip:\n pip install git+https://github.com/54yyyu/kaggle-mcp.git\n\n- Using uv (preferred for some environments):\n uv pip install git+https://github.com/54yyyu/kaggle-mcp.git\n\n3) Verify installation: ensure the executable or module kaggle_mcp is available. You should be able to run the MCP server using the configured command (see mcp_config).\n\n4) Configure Kaggle API credentials as described in the documentation (kaggle.json in ~/.kaggle or via Claude authentication).

Tips and common issues:\n- Ensure your Kaggle API credentials are present at ~/.kaggle/kaggle.json with correct permissions (chmod 600).\n- If you move the kaggle.json file, update KAGGLE_CONFIG_DIR accordingly.\n- The MCP server relies on the Kaggle API; make sure your environment has network access to Kaggle.\n- If you encounter authentication errors, re-create the Kaggle API token from your Kaggle account and replace kaggle.json.\n- For best results, keep Python 3.8+ and the MCP SDK up to date.\n- When running through Claude, you may opt to authenticate using the authenticate() tool with your username and API key if prompted.\n- If you use uv, ensure it is correctly installed and that the pip package is installed within the uv environment.\n- The configuration example to Claude Desktop shows a minimal setup; you can extend mcpServers with additional options or environment variables as needed.