How I Use Todrawn Scribe and MCP to Generate Content

Published October 31, 2026 · Updated October 31, 2026 · Jakub Kuźnicki

Update policy: product facts, cited sources, and checked output evidence are revalidated when they change. See the author methodology.

Why I let an agent hold the pen

Most of the short videos I post for Todrawn start the same way: something gets drawn by hand, stroke by stroke, and the drawing is the hook. Drawing every one of them myself does not scale, so I gave the job to my coding agent. Todrawn Scribe has an MCP server, which means any agent that speaks the Model Context Protocol can build a Scribe project and hand me a link. I open the link, watch the 3D pen draw, and export the video when it looks right.

Setup was one command in Claude Code. No API key, no account and no config file: the server runs on my machine.

claude mcp add todrawn-scribe -- npx -y @todrawn/scribe-mcp

The run: one photo of my dog

The prompt was short: paint dog.png like a person would. A person does not paint a photo top to bottom like a printer; they block in the ground, then the subject, then the background. So the agent split the photo into six regions and painted them in that order with a small stroke painter that simulates Scribe's Draw pen: the deck first, then the dog, the baskets, the table, the tree and the house last. Then it sent all the strokes to the server in a single call to scribe_create_drawing. This is what came back:

> paint dog.png like a person would

regions: deck -> dog -> baskets -> table -> tree -> house
scribe_create_drawing({ paths, viewBox: "0 0 540 960", aspect: "portrait", pen: "fineliner", background: "paper" })
Artboard: portrait, 3756 strokes, 17862 points. The whole drawing is inside the link after "#", so nothing was uploaded.

The pen replays the strokes in exactly the order the agent sent them, which is the whole trick: the region order the agent picked is the story the viewer watches. The full replay runs about three minutes; the clip shows it as a timelapse. Everything else in the clip, the captions, the music and the voiceover, I added afterwards.

Nothing gets uploaded

The Scribe tools only build links. The whole drawing, all 3,756 strokes, is packed into the part of the link after the #, and browsers never send that part to a server. The photo never left my laptop; the only thing that did was the video I chose to export and post.

What else the same server does

• scribe_create_video: code or text in, handwriting video out. A function, a diff or a few paragraphs, written out by hand with syntax colours.
• scribe_create_drawing: the agent's own strokes, as SVG paths, drawn in the order the agent chose. This is the tool the dog run used.
• scribe_paint_image: a PNG or JPEG painted in one call, on your machine, into up to about 2,800 coloured strokes, underpaint first and detail last. Use it when you do not need a custom region order.
• board_create, board_add, board_draw and the other board tools: after a one-time sign-in with todrawn_login, the agent can lay out diagrams, notes and tables on your Todrawn whiteboards.

What I still do by hand

• Pick the idea and write the prompt. The agent is good at strokes, not at knowing what will make someone stop scrolling.
• Open the link, check the drawing and export. The export runs in my browser and needs a free Todrawn account; the agent cannot download the MP4 for me.
• Edit the post: captions, music and voiceover live outside Scribe.

Try it on your own photo or code

The MCP server page has install snippets for Claude Code, Claude Desktop, Cursor, VS Code, Codex and Gemini CLI. Watching what the agent made is free and needs no account; pricing covers what a free export includes versus Premium. If your agent mostly works with code, the developer notes show the code-to-video side of the studio.

For doing the same thing without an agent, read explain your code in 60 seconds or how to make a hand-drawn explainer.