@unovis/mcp documentation

An MCP server that turns data into Unovis charts — rendered locally in Node, with no browser and no remote services.

Pages

Page	What's in it
Getting started	Install and connect from Claude Code, Claude Desktop, Cursor, Codex, VS Code; HTTP transport; environment variables
Tools reference	Every tool with all its options, generated from the live schemas
Output types	svg, png, html, interactive, config, code — what each returns and when to use it
Chart spec	The JSON IR behind every chart: containers, components, accessor references
Programmatic use	Use it as a library: renderChart, the renderToSvg SSR primitive, PNG rasterization, embedding the tools in your own MCP server
Interactive charts	The browser widget, self-contained HTML, iframe embedding, the MCP Apps extension
Native WebViews	React Native embedding: the bridge, theme switching without reloads, version pinning
Architecture	How headless rendering actually works, and why each piece exists
Troubleshooting	Fonts, blank charts, timeouts, large data, known limitations

The 30-second version

claude mcp add unovis -- npx -y @unovis/mcp

Then ask for a chart. The agent calls a tool with your data:

{
  "name": "generate_line_chart",
  "arguments": {
    "data": [
      { "month": "2024-01", "sales": 120, "cost": 80 },
      { "month": "2024-02", "sales": 150, "cost": 95 }
    ],
    "x": "month",
    "xIsTime": true,
    "y": ["sales", "cost"],
    "seriesLabels": ["Sales", "Cost"],
    "title": "Revenue"
  }
}

…and gets back a standalone SVG. Change outputType and the same call returns a PNG, an interactive HTML file, the chart spec as JSON, or ready-to-paste React/Svelte/Vue/Angular/Solid source.

What makes it different

• Everything is local. Charts render in-process with jsdom; your data never leaves the machine, and nothing needs a network connection after install.
• One spec, many outputs. A single JSON chart spec drives the headless renderer, the interactive browser widget, and the code generator.
• It's also a library. renderChart, renderToSvg and the tool registry are exported, so the same engine works in scripts, CI, and your own MCP server.