Composable dashboards (Private Preview)

ℹ️

Private preview

Composable dashboards and the metrics API are in private preview: available to invited accounts, and features may change. See Integration and feature lifecycles for what this means, and reach out to your account representative to request access.

Overview

Ask your AI agent about your data in Monte Carlo (data quality, alerts and incidents, monitoring coverage, credits and monitor runs) and it builds the dashboard for you, sliced by whatever dimensions you need.

How it works

1. Connect once. Add the Monte Carlo MCP server to your AI tool. In a coding agent (Claude Code, Cursor, Codex, GitHub Copilot), also install the dashboard skills from the Monte Carlo agent toolkit: they teach the agent the layouts and queries on these pages. In Claude (desktop and web) and ChatGPT, add Monte Carlo as a connector; the dashboard skills are not available there yet, so the agent works from the metrics' own descriptions. This is a one-time setup.
2. Ask. Ask a question in your own words, or start from a ready-made prompt in Dashboard examples. For example:
   • "Build a report on data quality for my Payments data product."
   • "How many alerts did we get per week over the last 3 months, and how fast did we fix them?"
   • "Which key tables have no monitor?"
   • "Which monitor types used the most credits this quarter?"
3. Get the dashboard. Your agent builds it in your AI tool. Every number comes from the same metric definitions, so answers stay consistent whichever AI tool you use.

Recommended AI models

📘

Use a mid-tier or top-tier model

Composable dashboards work best with current mid-tier and top-tier models, such as Claude Sonnet or Opus, or GPT gpt-6-astra or gpt-6.1-sol. Small and low-cost models often misread the values they get back and can show numbers that are not in your data.

The metrics API returns the same, correct numbers to every agent. What ends up on your dashboard depends on what the model does with them: whether it picks the right metrics, reads the matching example and Dashboard style, and puts the answers on the page as they are.

Capable models do all three. Smaller models tend to skip these pages, misread what a value means, distort numbers when they copy or add them up, and fill gaps with estimates. Asking them in the prompt to read the docs does not reliably change that, and the result can look complete and still be wrong.

We have done our best to keep every agent on track: each metric says what it means, and the examples and Dashboard style show how to query and present it. We cannot fully guarantee what a model does with the answers, so the choice of model matters.

In October 2026 we tested the dashboard examples and common questions on these models, checking every number on the page against the data:

Whichever model you use, make sure your agent has these pages: install the dashboard skills where your AI tool supports them, or point the agent to Dashboard examples and Dashboard style.

📘

Data scope and freshness

Your agent only sees your account, and only the domains you have access to. The data is refreshed periodically, and every answer states when it was last refreshed. Each metrics page says how far back its metrics go and what it does not cover; for periods, see Time in the Glossary.

Building it yourself?

If you are building in a BI tool or writing your own queries, everything the agent uses is documented here:

1. How to query: how to send a query to the metrics API, the parts of a query, names, time, filters and limits.
2. Data catalog: every metric and dimension, and the product words they map to.
3. Dashboard examples: ready-made dashboards with their prompt, layout and queries.
4. Dashboard style: how a dashboard should look, in any tool.