Set it up

In datakumi, open Settings › API tokens and create a token For an AI assistant (tick Read only for answers only). Web AI apps don't need this: they connect to the hosted server at https://<datakumi>/mcp and sign in (ai-assistant.md, "A hosted MCP server").
Install the server's dependencies (Python 3.10 or later) in the datakumi folder:
```
pip install -r mdm_mcp/requirements.txt
```
Add it to your AI app. It's started with python -m mdm_mcp, with two settings: MDM_API_URL (where datakumi is, e.g. https://app.datakumi.com or http://localhost:8000) and MDM_API_KEY (the token).

Claude Desktop (Settings › Developer › Edit config, claude_desktop_config.json), Cursor (.cursor/mcp.json) and Windsurf use the same shape:

{
  "mcpServers": {
    "datakumi": {
      "command": "python",
      "args": ["-m", "mdm_mcp"],
      "cwd": "C:/path/to/mdm",
      "env": { "MDM_API_URL": "http://localhost:8000", "MDM_API_KEY": "mdm_..." }
    }
  }
}

Claude Code:

claude mcp add datakumi --env MDM_API_URL=http://localhost:8000 --env MDM_API_KEY=mdm_... -- python -m mdm_mcp

VS Code (.vscode/mcp.json): {"servers": {"datakumi": {"type": "stdio", "command": "python", "args": ["-m", "mdm_mcp"], "env": {...}}}}

Local models. Run the model with Ollama, LM Studio or vLLM, and connect it through an MCP-capable chat client (Open WebUI, LibreChat, Continue) using the same command and settings. Pick a model that's good at tool calling.