Cross-Vendor Comparison Matrix¶
One screen for the whole landscape. COVERAGE.md groups families by paradigm (so you learn a stack
once and cover a cluster); this page is the orthogonal view — every family in one flat table, column-by-column, so
you can eyeball how any two protocols differ on transport, discovery, auth, push/poll, progress semantics, and
multi-material.
This is a distilled cross-index, not the source of truth. Every cell is a compression of a per-family paper in
protocols/ — when a cell and a paper disagree, the paper wins. Cells are deliberately
terse; follow the family link for the wire strings, gotchas, and confidence gaps. — means the paper documents no such
facet (not "unknown"); "see paper" means the answer is real but too branchy for a cell.
Confidence tags: 🟢 hardware-validated · 🟡 source-read · 🔵 community/RE · ⚪ inferred. Only Anycubic and Bambu carry hardware validation; every other family is 🟡 source-read (wire-shape from the vendor's own slicer/SDK/docs, not yet confirmed on a bench). Trust the per-paper tags on individual facts over this summary column.
Machine-readable: a structured mirror of this matrix — per-family transport/ports/auth/model/progress/feeder/confidence plus the fingerprint routing — lives at
data/comparison.json, regenerated from this page.
Main matrix¶
| Family | Paradigm | Transport (proto:port) | Discovery | Auth / credential | Push vs Poll | Progress semantics | Multi-material | Conf. |
|---|---|---|---|---|---|---|---|---|
| Anycubic | Proprietary LAN | MQTT/TLS :9883 + HTTP :18910 (identity/upload) |
manual IP; GET :18910/info (sleeping → HTTP-silent) |
slicer mTLS client cert + MQTT creds derived from /info |
poll-hybrid (idle pushes nothing ~48 s; active auto-pushes) | % + reported times |
ACE per-slot (multiColorBox) + external spool |
🟢 |
| Bambu | Proprietary LAN | MQTT/TLS :8883 + implicit FTPS :990 |
SSDP :2021; manual IP |
user bblp, password = access code off the printer screen |
push (pushall snapshot → live deltas); no command ack |
% + time |
AMS (full/Lite/high-temp) per-slot + external | 🟢 / 🟡 |
| Creality (stock OS) | Proprietary LAN | plain WS ws://:9999 + HTTP :80/upload |
manual IP; vendor UDP broadcast; GET /info |
none on LAN socket (LAN-trust) | push-hybrid (unsolicited state merged; get poll backstop) |
% — treat as file-byte (Klipper under the hood) |
CFS per-slot, rich (per-slot % + RFID + dry-box) | 🟡 |
| Elegoo | Mixed (3 stacks) | Moonraker :7125 / SDCP WS :3030 / MQTT :1883 + HTTP :80 |
manual; SSDP·mDNS / UDP :3000 / UDP :52700 |
none / none (LAN-trust) / access code (CC2) | subscribe / passive push + poll floor / push + 1002 poll |
see paper (per-backend) | CANVAS 4-slot (material/color/temp; no qty, no RFID UID) | 🟡 / 🔵 |
| Klipper / Moonraker | Moonraker | HTTP + WS JSON-RPC :7125 |
manual IP baseline; optional mDNS _moonraker._tcp / SSDP |
often none (trusted_clients); else API key (X-Api-Key) / JWT |
subscribe (WS): full snapshot → sparse deltas; or one-shot query | file-byte position (not time) | none native — a software provider (Happy Hare / AFC) layers it | 🟡 |
| Snapmaker U1 | Moonraker + wrapper | Moonraker :7125 + HTTP :8100/api (pair/orchestrate) |
manual IP; fingerprint by probing :8100/api |
pairing: authCode (presence) + accessCode (session); link_mode lan/wan |
subscribe (WS) + poll fallback | file-byte; time in seconds | toolchanger (filament→tool map), not a feeder | 🟡 |
| Duet3D / RRF | Object-model | HTTP :80 — standalone rr_* or SBC/DSF /machine/* REST + 1 WS |
manual host/IP; <name>.local mDNS/NetBIOS where honoured |
machine password → session → X-Session-Key header |
standalone poll-only (no push); SBC push (full-model → patch) | file-byte fraction (state value is processing) |
none (no MMU model) — toolchanger tools[] array |
🟡 |
| OctoPrint | Host controller | HTTP REST /api/* :5000 + SockJS /sockjs |
mDNS _octoprint._tcp; manual IP/URL |
one API key (X-Api-Key / Bearer); Application-Keys handshake |
push-primary (SockJS current/history) + REST poll backstop |
fraction 0.0–1.0, file-byte (not time) |
none in core — exposes multi-tool; feeders are plugins | 🟡 |
| PrusaLink | Host controller | HTTP :80 only — no push channel |
manual IP; mDNS advertised (service type unconfirmed); GET /api/v1/info |
HTTP Digest (user maker); legacy X-Api-Key tier (off by default) |
poll-only (GET /api/v1/status) |
job.progress time-based; seconds |
presence only (info.mmu bare boolean) |
🟡 |
| FlashForge | Proprietary LAN | newer: pub/sub SDK (closed DLL, MQTT-shaped) / legacy: raw-TCP G-code :8899 |
newer: UDP broadcast (per-device port); legacy: manual IP → :8899 |
newer: per-device checkCode (screen); legacy: ~M601 handshake, no code |
newer: subscribe + query; legacy: poll (~M119/~M105/~M27) |
see paper (Klipper inside newer; SD-byte on legacy) | Material Station 4-spool (AD5X, newer path only) | 🟡 / 🔵 |
| Marlin (USB serial) | USB serial | USB-CDC virtual serial (COMx / ttyUSB* / ttyACM*); baud autodetect |
none (OS port enum + M115 probe); VID/PID = UART chip, not printer |
none — physical access is the authorization | poll-only (M105/M114/M27; M155 opt-in auto-report) |
synthesized from M27 SD byte position (host-stream: line count) |
none queryable (T<n> lives in the sliced file) |
🟡 |
Pick your paradigm¶
Route by the live service you observe, never by the model name (several vendors span multiple stacks; Creality and Elegoo each span three). Fingerprint on connect and re-fingerprint on reconnect.
| You observe… | It's probably… | Paper |
|---|---|---|
Port :7125 answering JSON-RPC 2.0; printer.objects.list works |
generic Moonraker / Klipper (or a vendor fork) | klipper-moonraker.md |
A Moonraker host that also answers :8100/api |
Snapmaker U1 (Moonraker behind a pairing wrapper) | snapmaker.md |
MQTT on :8883 wanting user bblp + an access code; SSDP on :2021 |
Bambu Lab | bambu.md |
MQTT on :9883 needing a slicer mTLS cert; GET :18910/info returns identity |
Anycubic (Kobra) | anycubic.md |
A plain ws://:9999 carrying flat {method,params}, no auth; :80/upload |
Creality stock OS | creality.md |
JSON-over-WebSocket on :3030 with topic strings; UDP discovery on :3000 |
Elegoo SDCP (Centauri Carbon 1) | elegoo.md |
A printer running its own MQTT broker on :1883 + :80 upload + access code |
Elegoo CC2 | elegoo.md |
GET /rr_model (or /rr_connect) returns JSON; X-Session-Key after login |
Duet / RepRapFirmware (standalone) | duet.md |
A /machine/* REST API + one WebSocket at /machine |
Duet / RRF on an SBC (DSF) | duet.md |
/api/* with an X-Api-Key header + a SockJS /sockjs push channel |
OctoPrint host | octoprint.md |
GET /api/v1/status over HTTP Digest (user maker); no push channel |
PrusaLink host | prusalink.md |
A device answering ASCII G-code on raw TCP :8899 (or UDP SDK discovery + checkCode) |
FlashForge (legacy / newer) | flashforge.md |
A USB serial port that emits a Marlin start banner and answers M115 |
Marlin over USB | marlin-serial.md |
Traps that bite everyone¶
The same four mistakes recur across families — each has a cross-cutting pattern write-up:
- Time units drift (minutes vs seconds vs milliseconds). Different fields carry different units; a single slip
turns an 11-minute print into 112 hours. Normalize to integer seconds at the adapter edge.
→
patterns/timing-normalization.md - Progress is file-byte position, not time. Most Klipper/Moonraker-descended and object-model stacks report a
byte fraction (Klipper, Creality, Duet, OctoPrint, Snapmaker); a few are genuinely time-based (PrusaLink). Naive
elapsed / progressETA is systematically wrong near the end for the byte kind. →patterns/timing-normalization.md - Push vs poll — synthesize the missing stream. Poll-only stacks (PrusaLink, Duet standalone, Marlin) have no
event channel; diff successive snapshots into a change-event stream so the rest of your code sees one shape.
→
patterns/discovery-and-credentials.md(§ poll-synth) - Feeder is presence, not a vendor string. Detect a multi-material unit by the capability it reports, not by a
brand name; DIY units are identified by their software provider (Happy Hare / AFC), and several vendor units omit
quantity or RFID UID entirely.
→
patterns/multi-material-feeders.md
Confidence footnote. The
Conf.column is a one-glyph summary of a whole family; individual facts inside each paper carry their own 🟢/🟡/🔵/⚪ tags, and those are authoritative. A 🟡 here means "wire-shape read from the vendor's own source, not yet confirmed on hardware" — promoting any 🟡 family to 🟢 with a real capture is the highest-value contribution (see each paper's Confidence & validation section for the exact open gaps). Sanitized throughout: example IPs use192.0.2.x(RFC 5737); credentials are described as mechanisms, never values.