Skip to main content
Vistazo is a dashboard tool with an agent where the query builder normally goes. You connect data sources as MCP servers, describe what you want to see, and the agent writes the chart: which source to call, how to reshape the rows, what kind of chart it should be. The result is a real dashboard. It refreshes on a schedule, it can be shared with a link, and every widget stays wired to the source it came from. There is no semantic layer to model and no SQL layer to maintain. The agent reads your sources’ own tool descriptions and works from those.

​
The shape of it

1

Connect a source

Anything that speaks MCP. Five public data sources are hosted for you and need no setup, and your own servers connect over OAuth where they require it. See Connections.
2

Ask in chat

Describe the chart. The agent calls your sources, shapes the data, and shows you a preview in the conversation before anything is saved. See Chat.
3

Add it to a dashboard

The preview becomes a widget. From there it refreshes on its own schedule, and you can ask for it to be moved, resized or restyled at any point. See Dashboards.

​
Two ways in

Vistazo is an MCP client, and it is also an MCP server. The same operations the in-app chat performs are available to your own agent.

Use the app

Sign in, look at your sources, build your first dashboard.

Drive it from an agent

Connect Claude Code, Cursor, Codex or any MCP client and build dashboards from your editor.

​
Things worth knowing early

  • You read in the app and build by asking. The canvas has no drag handles. Layout, styling and wiring are all set through the agent, which is also why they survive a rebuild.
  • Viewing is free and unlimited. No per-viewer charges, and no seat needed to read a dashboard or open a share link.
  • The dashboard owns the chart. Once you add a preview, you edit it on the dashboard rather than re-drafting it in chat. The conversation keeps showing that widget live, so dashboard changes appear in the thread, but nothing flows the other way.
  • Refreshes are cheap. A scheduled refresh re-runs the stored transform with no model call, which is why they are not reserved for paid plans. The model returns only when a source changes shape and the transform needs repairing.
  • Parameters are authored, not remembered. A filter you change while reading is yours and is never written back for everyone else. See Parameters.