Summary
Analytics you can talk to, and then change.
Most BI integrations stop at reading. This one creates datasources, queries, widgets, dashboards and alerts, so a question that turns out to be worth asking every week becomes a saved artifact in the same conversation.
What it is
The MCP endpoint built into Knowi itself — a hosted analytics platform — reached over Streamable HTTP at an endpoint on your own instance.
What you get
- A data question answered end to end: the dataset found, a query generated, results returned
- Search across every dashboard, widget, dataset and report when you do not know where a number lives
- Dashboards, widgets and cards created and laid out, not merely read
- Datasources and queries created, and data pushed into a dataset that is created if it does not exist
- Alerts created, listed, tested and updated
- Documents ingested, questioned directly, and their data extracted
- CSV and PDF exports, and an embed URL for putting a widget somewhere else
- Natural-language and deterministic tools kept separate, so you can choose predictability over interpretation
Requirements
An MCP token generated in Knowi under Settings then AI Settings — it is shown once. Add the endpoint with that token as a bearer header; there is nothing to install, since the server ships inside Knowi.
Setup effort
One command plus a key — claude mcp add knowi https://www.knowi.com/api/2.0/mcp \ --transport http \ --header "Authorization: Bearer YOUR_MCP_TOKEN", then supply credentials