mmdr

Render Mermaid diagrams from Python or the command line — fast, native, no browser required.

Most Mermaid rendering tools work by spinning up a headless browser and running the official JavaScript library. This works, but it's slow to start, heavy to install, and awkward to embed in a pipeline. mmdr takes a different approach: the rendering engine is written in Rust and ships as a compiled Python extension. Install it with pip, import it, and start rendering — no Node.js, no Puppeteer, no Chrome, no JS runtime of any kind.

pip install mmdr

That's it. No system dependencies, no extra steps.

Output formats: SVG · PNG · raw RGBA pixels · NumPy array

Two backends, one package:

• merman (default) — targets full parity with the official Mermaid @11.15.0 library, supports all diagram types including newer ones like kanban, architecture, and c4
• mermaid-rs-renderer — a lighter, faster alternative that covers the most common diagram types

---

CLI

mmdr ships with a command-line tool that works similarly to the official mermaid-cli (mmdc), but starts in milliseconds instead of seconds because there's no browser to boot.

Quick start

# render a file to SVG
mmdr -i diagram.mmd -o output.svg

# render to PNG
mmdr -i diagram.mmd -o output.png

# read from stdin, write to stdout — works great in pipelines
echo 'flowchart LR; A-->B-->C' | mmdr -i - -o -

# save the output to a file from a pipe
echo 'flowchart LR; A-->B-->C' | mmdr -i - -o diagram.svg

PNG options

mmdr -i diagram.mmd -o output.png \
  --width 1200 \               # canvas width in pixels
  --height 800 \               # canvas height in pixels
  --background '#ffffff'       # background color (transparent by default)

Backend and theme

# use the faster backend for simple diagrams
mmdr -i diagram.mmd -o output.svg --backend mermaid-rs-renderer

# classic Mermaid theme (mermaid-rs-renderer only)
mmdr -i diagram.mmd -o output.svg --backend mermaid-rs-renderer --theme classic

Batch rendering

# convert every .mmd file in a directory
for f in diagrams/*.mmd; do
    mmdr -i "$f" -o "${f%.mmd}.svg"
done

# same, but PNG with a white background
for f in diagrams/*.mmd; do
    mmdr -i "$f" -o "${f%.mmd}.png" --background '#ffffff'
done

Other commands

mmdr --info       # render Mermaid's built-in info diagram and show its text
mmdr --version    # print the mmdr version
mmdr -h           # full list of options

---

Python API

Rendering

mmdr.render() accepts a Mermaid source string and returns a Diagram object. The actual rendering is lazy — nothing runs until you ask for an output. The SVG result is cached, so calling .svg() multiple times only renders once.

import mmdr

d = mmdr.render("""
flowchart TD
    A[Start] --> B{Is it working?}
    B -- Yes --> C[Great!]
    B -- No  --> D[Debug it]
    D --> B
""")

SVG

svg = d.svg()   # str

# write to file
with open("output.svg", "w") as f:
    f.write(svg)

PNG

# default: transparent background, size determined by the diagram
png = d.png()

# with explicit size and background
png = d.png(width=1200, height=800, background="#ffffff")

with open("output.png", "wb") as f:
    f.write(png)

Save — auto-detect format

d.save("output.svg")                                  # writes SVG
d.save("output.png")                                  # writes PNG
d.save("output.png", width=1200, background="#ffffff")  # PNG with options

Raw pixels

.raw() returns the pixel buffer directly from the renderer as (bytes, width, height). The bytes are raw RGBA8888 data — 4 bytes per pixel, row-major, top-to-bottom. No encoding, no copying through an image library.

raw, w, h = d.raw(background="#ffffff")
print(f"image is {w}×{h} pixels, {len(raw)} bytes")

NumPy

.numpy() returns an (H, W, 4) array with dtype uint8 and RGBA channel order. It's built on .raw() — requires only numpy, no Pillow.

arr = d.numpy()
print(arr.shape)   # e.g. (480, 640, 4)
print(arr.dtype)   # uint8

# drop alpha → RGB
rgb = arr[:, :, :3]

# flip upside-down
import numpy as np
flipped = np.flipud(arr)

Jupyter / IPython

Diagram implements _repr_svg_(), so notebooks render it inline without any extra code:

import mmdr

# just evaluating this in a cell displays the diagram
mmdr.render("sequenceDiagram\n  Alice->>Bob: Hello!\n  Bob-->>Alice: Hi!")

---

Backends

mmdr.backends()
# ['merman', 'mermaid-rs-renderer']

# merman is the default — pick it explicitly if you want to be clear
d = mmdr.render("flowchart LR; A-->B", backend="merman")

# mermaid-rs-renderer for simpler diagrams where speed matters
d = mmdr.render(
    "flowchart LR; A-->B-->C",
    backend="mermaid-rs-renderer",
    theme="classic",      # "modern" (default) or "classic"
    node_spacing=60.0,
    rank_spacing=80.0,
    aspect_ratio=(16, 9),
)

The theme, node_spacing, rank_spacing, and aspect_ratio options only apply to mermaid-rs-renderer. The merman backend uses its own layout engine tuned for parity with the official library.

---

Low-level utilities

If you already have an SVG string from another source, you can rasterize it directly without going through render():

from mmdr import svg_to_png, svg_to_raw

svg = open("diagram.svg").read()

png = svg_to_png(svg, width=1200, background="#ffffff")
raw, w, h = svg_to_raw(svg)

---

Supported diagram types

Diagram type	merman	mermaid-rs-renderer
flowchart / graph	✅	✅
sequenceDiagram	✅	✅
classDiagram	✅	✅
stateDiagram	✅	✅
erDiagram	✅	✅
pie	✅	✅
gantt	✅	✅
timeline	✅	✅
mindmap	✅	✅
gitGraph	✅	✅
xychart	✅	✅
block diagram	✅	❌
architecture	✅	❌
kanban	✅	❌
c4diagram	✅	❌
sankey	✅	❌
packet	✅	❌

---

License

MIT