Comparison
Both let an AI agent work with Excalidraw. They solve different problems, and the honest answer to "which should I use" is "probably both, for different reasons." Here's the distinction.
The official Excalidraw MCP is the primitive layer — raw element CRUD: create a rectangle, move it, connect two shapes. excalidraw-skill-pack is the quality layer on top — the opinionated methodology that decides which elements, where, and why, so the result argues instead of labels.
| Official Excalidraw MCP | excalidraw-skill-pack | |
|---|---|---|
| What it is | An MCP server for element-level Excalidraw operations | A skill + MCP server encoding diagram methodology, themes, and a renderer |
| You control | Individual elements, directly | The intent; the methodology handles layout |
| Layout quality | Whatever you or the agent place | Heuristics enforce hierarchy, direction, grouping |
| Themes | — | Swappable, publishable theme packages |
| Reproducible render | — | Golden-fixture PNG rendering (Node + Python parity) |
| Best for | Direct, fine-grained element control | "Draw me a good diagram" without art-directing it |
| License | Open source | MIT |
You want to place and wire specific elements yourself, you're building your own higher-level tooling on top of Excalidraw primitives, or you don't want an opinion imposed on layout. It's the right primitive.
You want to ask an agent for an architecture, protocol, or state-machine diagram and get back a figure that teaches — with a visual hierarchy and a flow your eye can follow — instead of labeled boxes. It rendered 77 figures for a published engineering book, which is the honest test of whether the methodology holds up.
Yes — think of the official MCP as the hands and this as the taste. Install excalidraw-skill-pack in one line:
/plugin marketplace add isatimur/excalidraw-skill-pack
MIT licensed. Works in Claude Code, Cursor, Codex, and Gemini CLI.