printd

A self-hosted print daemon. Send it a model file or URL, and it slices with your profiles, enforces your rules, renders a preview for approval, starts the print, and watches it, sending phone notifications with live camera photos at every stage. Drive it from any MCP client (Claude, other agents, your own tools) or the bundled CLI.

How it works

submit (file or model URL)
  -> slice (your slicer, your profiles)
  -> checks (brim policy, bed placement, motion bounds)
  -> preview image -> approval (overridable per job)
  -> start (preheat, then print)
  -> watch (camera-photo notifications: first layer, milestones, stalls, done/failed)

The watcher never acts on its own: it photographs the print bed with your printer's camera and notifies you; pausing or cancelling stays a human decision.

Related MCP server: Klipper MCP Server

Suggested hardware setup

A single machine can play both server roles if it has the RAM to slice. Keeping them separate means nothing you add to the printd host can disturb a running print.

Quickstart

pip install printd
printd-server --config config.yaml   # MCP server (HTTP)
printd-watch --config config.yaml    # watcher daemon
printctl status                      # same operations from the shell

Copy example-config.yaml and fill in your printer. printd authenticates to OctoPrint with an API key because OctoPrint requires one for all REST calls; copy it from OctoPrint under Settings → API.

Previews render as a single composite PNG: six shaded 3D views (extrusion tubes, per-feature colors), six transparent line projections, and a bed placement map. This needs Node plus a headless Chromium (cd render3d && npm install && npm run build, then npx playwright install chromium or point PRINTD_CHROMIUM at any Chromium binary). Without Node, printd falls back to a five-view SVG line render that only needs Chromium; the pipeline works either way.

Swappable parts

Everything opinionated is configured, not hardcoded, in one YAML file:

License

MIT