A cozy shared craft room where a human and an AI can draw together, place perler beads one handful at a time, finish the work, and display it in a little gallery.
The project includes a mobile web UI and a Stateless Streamable HTTP MCP endpoint. It is a standalone craft app: there are no diary, chat-memory, or private relationship features.
Shared portrait or landscape sketchbook with live strokes, color presence, naturalized AI paths, brushes, fill, and actor-scoped undo
Perler boards in together, separate, and hidden gift modes
MARD color codes, per-participant bead inventories, reference patterns, resumable work plans, and 1–5 bead handfuls
Hands-on cover, heat, iron, cool, and lift workflow
Finished-work decoration with transparent PNG export, pouch presentation, outlines, and shadows
Shared gallery with picture frames, a tabletop stand, and size-aware hanging pouch slots
Participant-specific web and MCP access keys stored only as SHA-256 hashes
Generated anime-style room artwork bundled locally; no external image CDN is required
Related MCP server: pixel-mcp
Requires Python 3.11+.
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
export CRAFT_ADMIN_KEY="replace-with-a-long-random-secret"
shared-craft-roomOpen http://127.0.0.1:8000/setup/<CRAFT_ADMIN_KEY>/ and create at least two participants. The setup page returns a private room URL and MCP URL for each participant. An access key is shown only when it is created or rotated.
Variable | Default | Purpose |
| empty | Required secret for participant setup |
|
| Persistent JSON and PNG data directory |
| request origin | Public base URL returned by setup |
|
| Listen address |
|
| Listen port |
The app does not automatically load .env; inject variables through your shell or hosting platform.
Drawing: new_canvas, get_canvas_snapshot, select_canvas_color, draw_path, brush_stroke, fill_region, undo_canvas, export_image, queue_canvas_strokes.
Perler beads: list_bead_boards, new_bead_board, open_bead_board, select_bead_color, set_bead_work_plan, pin_bead_reference, upload_bead_reference, place_beads, continue_bead_work, take_bead, refill_beads, get_bead_snapshot, reveal_bead_gift, work_bead_finish. Use get_craft_job only when a client needs explicit completion status.
pin_bead_reference accepts bead coordinates, generates a standard pattern image, and also saves the resumable work plan. upload_bead_reference accepts an existing PNG/JPEG/WebP data URL under 1 MB and pins only that image; it does not infer bead coordinates from the picture. Human participants can upload the same formats from the reference-note control in the web UI.
The two queue tools are quota-friendly: one MCP call schedules multiple human-scale actions, while the server still publishes every stroke or five-bead handful separately for the live web UI.
Access keys live in the MCP URL rather than tool arguments. Treat room and MCP URLs as passwords and avoid logging or sharing them.
python3 -m pytest -qRun behind HTTPS and persist CRAFT_DATA_DIR. A minimal command is:
uvicorn shared_craft_room.server:app --host 0.0.0.0 --port 8000The bundled room backgrounds are WebP to keep first-load size small. User-created works remain PNG to preserve transparency.
This repository contains no production database, participant keys, private artwork, or server-specific launcher. The data/ directory is ignored by Git.
This server cannot be deployed
Online whiteboard your AI agent draws on: live shared canvases, solo or with your team.
Build, version, review, and export websites, web apps, and games from a conversation.
Design, save, and run outcome-aligned AI workflows and verifiers, with reliable image output.
Design canvas for AI assets: generate images, video, code and screens, and edit canvases.
