Databox already collects the numbers; the gap was that reading them meant building a dashboard first. Asking directly skips that, and the ingest side means an agent can put data in as well as take it out.
A remote MCP server run by Databox at `https://mcp.databox.com/mcp`. It connects the metrics already flowing through your Databox account — from whichever of its integrations you have connected — to any MCP client, and can create data sources and ingest data as well as read it.
- Every Databox account your login can reach, listed
- The data sources in an account, listed
- A new data source container created for organising datasets
- The datasets under a data source, listed
- A dataset created, with primary keys where the data needs them
- Data ingested into a dataset
- A data source deleted
A Databox account — authorisation happens in the client rather than through a config file, and there is no additional charge for Databox users. Add `https://mcp.databox.com/mcp` as an HTTP connector: Claude Desktop takes it in `claude_desktop_config.json` with `"type": "http"`, Claude Web takes it under Settings → Connectors → Add Custom Connector, and Cursor and n8n take the same URL.
One command — npx -y mcp-remote https://mcp.databox.com/mcp