mviz: Turn JSON Into Polished Reports with Claude

Making a single decent chart on a web page is more work than it should be. You pick a charting library, wire up the data, fight with axis formatting, add a legend, make it responsive, then do it all again for the next chart. For a one-off report that nobody will maintain, that effort rarely pays off.

mviz collapses that work into a declarative spec. You say what you want (a bar chart here, a big headline number there, a table with sparklines below) and the tool renders a complete report. Because the output is a static HTML file, there is nothing to deploy and nothing to keep running. You open it in a browser or print it to PDF and you are done.

You write a Markdown file with embedded JSON chart specs, optionally pointing at external JSON or CSV files for the underlying data. mviz reads that file and expands each spec into a rendered chart using Apache ECharts (an open-source JavaScript charting library), then assembles everything into one HTML report with a light/dark theme toggle and print-to-PDF styling.

The AI-native part matters here. The spec format is compact and predictable, which is exactly the kind of thing a model like Claude is good at producing. So the realistic workflow is not "learn the spec syntax and hand-write JSON". It is "tell Claude what you want to see, let it write the spec, run mviz". That is why the project ships as a Claude skill rather than just a library.

mviz needs Node.js 24.12 or newer. There is no account to create and no API key to wire up for the tool itself. To add it to Claude Code as a skill:

npx skills add matsonj/mviz

To run it directly from the command line against a Markdown file:

npx mviz dashboard.md -o output.html   # generate the HTML report
npx mviz --lint dashboard.md           # validate the spec only

If you prefer Claude on the web, you download the skill file and add it to your project knowledge. Either way, the heavy lifting of writing the spec is meant to be handled by the model, not you.

For a tool this small, the range is surprisingly wide. mviz supports 17 chart types, including the usual bar, line, area, pie and scatter charts plus less common ones like heatmaps, sankey diagrams, waterfalls and bubble charts. On top of charts it offers 8 UI components such as big headline values, delta indicators, tables with inline sparklines, alerts and notes.

Layout is handled by a 16-column grid, so you can arrange a proper report rather than a single stacked column of charts. Number formatting auto-detects based on field names, which removes another fiddly step. The result prints cleanly to PDF, which makes it genuinely useful for sharing a snapshot with someone who will never open your code.

mviz fits anyone who occasionally needs a clean, shareable report and does not want to maintain a dashboard product to get one. Analysts exploring a dataset, developers documenting a result, or anyone already living in Claude Code who wants visual output will get value quickly. Because the artifact is a single static file, it is also a tidy way to archive a point-in-time view.

It is not a live dashboard. There is no auto-refreshing data, no user accounts, and no server-side anything by design. If you need a shared, always-current operational dashboard with permissions, this is the wrong tool and a hosted BI product is the right one. mviz is for fast, static, AI-assisted reports, and it stays deliberately within those lines.

mviz is a neat example of where AI-native tooling is heading: a tiny, focused tool that does one job well and leans on a model to handle the boilerplate. If your reporting needs are occasional and static rather than live and shared, it turns what used to be an afternoon of chart-wrangling into a one-line command and a short conversation with Claude. For the price (free) and the footprint (a single static file), it is an easy thing to keep in your back pocket.