// the find
mckinsey/vizro
Vizro is a low-code toolkit for building high-quality data visualization apps.
Vizro is a Python dashboard framework from McKinsey that wraps Plotly/Dash behind a declarative config layer — you describe your dashboard in Pydantic models, YAML, or JSON, and it handles layout, styling, and wiring. It targets data scientists and analysts who want production-looking internal tools without fighting Dash callbacks directly.
Pydantic as the config schema is a genuinely good call — you get validation, IDE autocomplete, and serialization to YAML/JSON for free, which makes dashboards version-controllable and LLM-friendly. The MCP server (Vizro-MCP) is a practical addition: letting Claude or Cursor generate dashboard configs rather than raw Dash code sidesteps most of the awkward API surface. The built-in design system means charts and layouts look consistent without any CSS work — McKinsey clearly dogfoods this and the visual quality shows. Active development (pushed yesterday) and a full CI pipeline with screenshot regression tests suggest this isn't abandonware.
Vizro-AI's dashboard generation has already been deprecated in favor of Vizro-MCP after less than a year — not a great sign for API stability if you built workflows on top of it. Anything outside the component model requires dropping to raw Dash, which means you're now maintaining two mental models simultaneously; the escape hatch is there but it leaks complexity. The framework abstracts Dash callbacks away entirely, which is convenient until something goes wrong and you're debugging through two layers of indirection with no clear place to look. No real backend data layer — you're still wiring up your own data loading; Vizro handles presentation only, so 'production-ready' depends heavily on what you bring to it.