Fetching the syntax help first is what makes this reliable: with the DSL rules in context, the assistant writes a diagram that renders on the first attempt rather than one that needs three corrections. Because the result is a live chhart.app URL rather than an image, the diagram stays editable after the conversation ends — which is usually what you want when a process description is going to change.
A generator for two diagram types on chhart.app. You write the diagram in a text DSL — or have the assistant write it — and the call returns a shareable URL to view and keep editing it.
- A flowchart built from the text DSL, with a title, returned as a link anyone you send it to can open
- A Sankey diagram from flow data, for the questions about where volume goes rather than what happens in what order
- The syntax documentation and worked examples for either diagram type, so the assistant writes valid DSL instead of guessing at it
Node.js 20 or higher for the local path, which runs as npx -y chhart-mcp. A public endpoint at https://mcp.chhart.app/mcp needs nothing installed at all, and the same server can be self-hosted over Streamable HTTP on PORT, default 3000. Either way the diagram is created on chhart.app and the result is a URL there, so the content leaves your machine and the link is what you share. MIT.
One command — npx -y chhart-mcp