Mermaid Graph Renderer
Renders Mermaid diagrams to production-quality SVG, PNG, and PDF in web and offline contexts. Covers the full spectrum from lazy-loaded client-side rendering to 1000x-faster native Rust batch export.
Core Constraint
Mermaid.js requires browser APIs (document.createElement , SVGTextElement.getBBox() ) to compute text dimensions. Every non-browser rendering path must either embed a headless browser or reimplement the renderer from scratch. This constraint shapes every decision below.
When to Use
✅ Use for:
-
Rendering Mermaid diagrams on web pages (performance-optimized)
-
Exporting diagrams to SVG, PNG, or PDF (offline/CLI)
-
Setting up CI/CD pipelines that generate diagram images
-
Choosing between rendering libraries for a project
-
Optimizing Mermaid load time on documentation sites
❌ NOT for:
-
Writing Mermaid syntax (use mermaid-graph-writer )
-
Rendering non-Mermaid formats (PlantUML, GraphViz — see Kroki)
-
General image processing or manipulation
Rendering Decision Tree
flowchart TD A{Where are you rendering?} -->|Browser| B{Performance matters?} A -->|CLI / CI| C{Volume?} A -->|Server-side| D{Must avoid client JS?}
B -->|Yes| E[Lazy-load mermaid.js] B -->|No| F[CDN script tag]
C -->|1-10 diagrams| G[mermaid-cli] C -->|10-100 diagrams| H{Need multiple formats?} C -->|100+ diagrams| I[mmdr Rust renderer]
H -->|Mermaid only| G H -->|Multiple languages| J[Kroki]
D -->|Yes, zero JS| K[Build-time with Puppeteer] D -->|Some JS OK| E K -->|Slow builds warning| L[rehype-mermaid + Playwright]
Web Rendering
Option 1: Lazy-Loaded mermaid.js (Recommended)
Best for documentation sites, blogs, and apps where only some pages have diagrams.
How it works: Check for .mermaid elements before loading the library. The ~480 KB cost is only paid on pages that actually need it.
// Lazy load pattern — only loads mermaid.js when diagrams exist function initMermaid() { const diagrams = document.querySelectorAll('.mermaid'); if (diagrams.length === 0) return;
const script = document.createElement('script'); script.src = 'https://cdn.jsdelivr.net/npm/mermaid@11/dist/mermaid.min.js'; script.onload = () => { mermaid.initialize({ startOnLoad: false, theme: 'neutral' }); mermaid.run({ nodes: diagrams }); }; document.head.appendChild(script); }
// Handle initial load + SPA navigation document.addEventListener('DOMContentLoaded', initMermaid); // For SPA route changes: call initMermaid() after navigation
Key detail: Handle both initial page load AND client-side navigation (SPA route changes) by re-running mermaid.run() when new diagram content appears.
Aspect Value
Bundle size ~480 KB (core + lazy diagram chunks)
Render time ~50-200 ms per diagram (client)
Diagram coverage Full (all Mermaid types)
Best for Docs sites, blogs, apps with occasional diagrams
Option 2: @mermaid-js/tiny
Pre-bundled subset for CDN use. All diagram types included upfront (no lazy-loading of diagram chunks). Smaller total but loaded eagerly.
Use when: You know exactly which 1-2 diagram types you need and want a single-file import.
Option 3: Build-Time SSR (Usually Not Worth It)
Render SVGs at build time using Puppeteer/Playwright. Sounds appealing; painful in practice.
Costs:
-
Adds ~280 MB to node_modules (Puppeteer + Chromium)
-
~2-3 seconds per diagram to spin up and render
-
A 5-second build becomes 45 seconds for ~30 diagrams
-
Docker/CI headaches (Chrome needs to be installed)
When actually worth it: Strict zero-JS requirements, or sites with hundreds of diagrams where you want to eliminate client-side rendering entirely.
Tools: rehype-mermaid (uses Playwright), rehype-mermaid-cli (wraps official CLI).
Verdict: Unless you have strict zero-JS requirements, lazy-loaded client-side rendering is the better trade-off.
Offline / CLI Rendering
Option 1: mermaid-cli (Official)
The canonical CLI tool. Launches headless Chromium, renders, captures output.
Install
npm install -g @mermaid-js/mermaid-cli
Basic usage
mmdc -i input.mmd -o output.svg mmdc -i input.mmd -o output.png -b transparent mmdc -i input.mmd -o output.pdf
With config
mmdc -i input.mmd -o output.svg -t dark --configFile mermaid.config.json
Batch: process Markdown with embedded diagrams
mmdc -i document.md -o document-with-images.md
Docker
docker run --rm -v $(pwd):/data minlag/mermaid-cli -i /data/input.mmd -o /data/output.svg
Aspect Value
Speed ~3000 ms per diagram (Chromium overhead)
Output SVG, PNG, PDF
Size ~280 MB (Puppeteer + Chromium)
Coverage Full (all Mermaid types, 100% parity)
Best for 1-50 diagrams, when accuracy matters most
SVG gotcha: Output SVGs contain <foreignObject> elements that break in Inkscape and rsvg-convert . Fix: set "htmlLabels": false in config, or export to PNG instead.
Option 2: Kroki (Multi-Format Gateway)
Unified HTTP API that routes diagram source to appropriate rendering engine. Supports 25+ languages (Mermaid, PlantUML, GraphViz, D2, Excalidraw, etc.).
Render via HTTP (self-hosted or kroki.io)
curl -X POST https://kroki.io/mermaid/svg
-H 'Content-Type: text/plain'
-d 'flowchart LR
A --> B --> C'
-o diagram.svg
Self-host via Docker Compose
docker-compose.yml needs: yuzutech/kroki + yuzutech/kroki-mermaid
Aspect Value
Speed ~500-2000 ms (network + render)
Output SVG, PNG, PDF, JPEG
Languages 25+ (Mermaid, PlantUML, GraphViz, D2, etc.)
Best for Multi-language pipelines, CI/CD, documentation generators
When Kroki wins: You need multiple diagram languages in one pipeline. When it doesn't: Simple "render one Mermaid diagram" — overkill.
Option 3: mmdr (Native Rust — Fastest)
Native Rust reimplementation. No browser, no Node.js. ~1000x faster than mermaid-cli.
Install (Rust toolchain required)
cargo install mmdr
Usage
mmdr input.mmd -o output.svg mmdr input.mmd -o output.png # via resvg, no browser
Aspect Value
Speed ~3 ms per diagram (1000x faster)
Output SVG (native), PNG (via resvg)
Size ~5-10 MB binary
Coverage 13 diagram types (partial parity)
Best for Batch processing 100+ diagrams, CI/CD speed
Trade-off: Not all Mermaid features are supported yet. Flowcharts and sequence diagrams are mature; other types vary. Accept imperfection for speed, or fall back to mermaid-cli for edge cases.
No <foreignObject> issue: Uses native SVG text, so output works with Inkscape and other SVG tools.
Comparison Matrix
Criterion mermaid.js CDN mermaid-cli Kroki mmdr (Rust)
Runtime Browser Node + Puppeteer Docker Native binary
Speed/diagram ~100 ms ~3000 ms ~1000 ms ~3 ms
SVG quality Excellent Excellent Excellent Good (improving)
PNG quality N/A Good Good Good (resvg)
PDF export N/A Yes Yes Not yet
Install size ~480 KB ~280 MB ~1 GB (Docker) ~5-10 MB
Diagram coverage Full Full Full 13 types
CI/CD fit Poor Moderate Good Excellent
Multi-language No No Yes (25+) No
Recommendations by Use Case
Use Case Recommendation Why
Docs site with occasional diagrams Lazy-loaded mermaid.js Only loads on pages that need it
Next.js / SSR app Dynamic import, ssr: false
Don't fight build-time rendering
CI/CD generating static docs (Mermaid only) mmdr for speed, mermaid-cli for accuracy 1000x faster vs. 100% feature parity
CI/CD with multiple diagram languages Kroki One API for PlantUML + Mermaid + GraphViz
Batch export for print/PDF (100+ diagrams) mmdr + mermaid-cli fallback Fast for most, accurate for edge cases
Desktop apps (Electron/Obsidian/VS Code) Built-in mermaid.js Chromium already embedded
Anti-Patterns
SSR Obsession
Wrong: Spending 3 days setting up Puppeteer in CI to avoid 480 KB of client JS. Right: Lazy-load mermaid.js. The 480 KB is only fetched on pages with diagrams. Most users won't notice.
foreignObject Surprise
Wrong: Exporting SVGs from mermaid-cli and wondering why Inkscape can't open them. Right: Set "htmlLabels": false in config, or export to PNG directly.
One Tool for Everything
Wrong: Using mermaid-cli for 500 diagrams in CI (25 minutes of build time). Right: Use mmdr for batch, mermaid-cli for the few diagrams that need full parity.