Summary

A diagram the model can fix, because it is told when the syntax is wrong.

Validation is the quiet feature here. Without it a malformed diagram comes back as a broken picture; with it the model gets an error it can act on and tries again.

What it is

A diagram generator that renders Mermaid on the server side. Full Mermaid syntax is supported, and the render step validates the input so a model that gets the syntax wrong is told, rather than handed a broken image.

What you get

Every Mermaid feature and syntax, with `backgroundColor` and `theme` configurable so a model can produce rich style settings
Six output types: `base64`, `svg`, `mermaid`, `file` — which saves a PNG to disk for agents that work with files — and the remote-friendly `svg_url` and `png_url` that share through public mermaid.ink links
Syntax validation, which is what lets a model iterate over several rounds towards correct output and correct graphics
Three transports selected with `--transport`: `stdio` by default, `sse`, and `streamable`, with `--port` and `--endpoint` to place them

Requirements

Nothing to install for the common case — configure your client with `npx -y mcp-mermaid`, using the `cmd /c` prefix on Windows. For a long-running server, `npm install -g mcp-mermaid` then `mcp-mermaid -t sse` or `mcp-mermaid -t streamable`; the SSE endpoint defaults to `/sse` on port 3033 and the streamable endpoint to `/mcp`, with the local development scripts using 1122 for streamable. A Docker image, `susuperli/mcp-mermaid:latest`, is published for both transports. It is also reachable through hosted platforms including aliyun, modelscope, glama.ai and smithery.ai over HTTP or SSE. MIT licensed.

Setup effort

One command — npm install -g mcp-mermaid