Bitfocus Companion’s new surface plugin system (shipped in 4.3.0) makes it possible to add support for new control surfaces without touching Companion itself. I used that to plug in the Ulanzi Stream Controller D200, a 13-button, 5×3 grid Stream-Deck-alike that’s cheaper than the Elgato but wasn’t supported out of the box.

Project: github.com/jcalado/companion-surface-d200

What it does

• 13 Companion-controlled buttons, each with a 196×196 icon
• Brightness, button events, press/release
• Small-window status display (clock / system stats), selectable per surface
• Drop-in installable as a developer module in Companion 4.3.0+

The reverse-engineering rabbit hole

redphx/strmdck had the wire protocol in Python: 1024-byte HID packets framed as 0x7c 0x7c [cmd:u16] [length:u32] [data...], with full button uploads pushed as ZIP archives containing a JSON manifest plus per-button PNGs. That gave me a head start, but a few things were out of date or wrong:

• strmdck targets VID/PID 2207:0019. On my Linux box, the D200 enumerated as 18d1:d002, Google’s ADB-style VID/PID. Standard USB control transfers to it failed with EPROTO, adb devices hit the same wall.
• My first port used node-hid, which couldn’t see the device at all. Switching to node-usb and bulk transfers got further but still failed with LIBUSB_TRANSFER_ERROR.

A USBPcap capture of Ulanzi Studio on Windows cleared everything up: the D200 actually exposes two USB devices simultaneously through an internal hub. The 18d1:d002 one is an abandoned/bootloader interface, dead on arrival from the host’s point of view. The real device is 2207:0019, a proper HID composite with two interrupt endpoints at wMaxPacketSize = 1024 plus a keyboard interface for the standalone hotkey buttons.

Windows enumerates both cleanly. Linux, connected directly, gives up after three tries:

usb 3-1: can't read configurations, error -22
usb 3-1: can't read configurations, error -22
usb usb3-port1: attempt power cycle
usb 3-1: New USB device found, idVendor=18d1, idProduct=d002

Only the useless one survives. The fix turned out to be stupidly simple: a USB 2.0 hub between the D200 and the host. With the extra hub layer, both child devices enumerate, usbhid binds to the HID interface, and node-hid just works.

That one detail burned a couple hours.

The ZIP manifest was wrong in strmdck too

The captured ZIPs from Ulanzi Studio use:

manifest.json
Images/<uuid>.png

(flat, no page/ prefix) and require a per-button Font object inside each ViewParam[0]. Getting icons to render required matching that structure exactly, including declaring the small-window slot 3_2 (which spans two cells) with SmallViewMode: 1.

Plus a lovely firmware bug: the byte at file offsets 1016 + 1024·N in the ZIP payload must not be 0x00 or 0x7c, or the upload is silently corrupted. strmdck’s workaround (retry compression with a random-length dummy file appended) ports over directly.

Writeup

The full wire protocol and ZIP format are documented in PROTOCOL.md; Linux-specific setup (hub requirement, udev rule, the diagnosis of the enumeration failure) is in SETUP.md.

Support

If this is useful to you: ko-fi.com/jcalado.