Control-only scrcpy-server process does not handle control messages (scid socket opened but no effect)

[vinhpdt55]

Hi, I’m running into a limitation when trying to run a control-only instance of scrcpy-server. Here’s what I’m doing:

• I push scrcpy-server.jar version 3.2 to /data/local/tmp/
• I launch two separate processes:

1. Video streaming (scid=0):

adb shell CLASSPATH=/data/local/tmp/scrcpy-server.jar \
app_process / com.genymobile.scrcpy.Server 3.2 \
video=true audio=false control=false cleanup=false tunnel_forward=true

2. Control-only process (scid=1):

adb shell CLASSPATH=/data/local/tmp/scrcpy-server.jar \
app_process / com.genymobile.scrcpy.Server 3.2 \
video=false audio=false control=true cleanup=false tunnel_forward=true scid=1

Then I forward the control socket:

adb forward tcp:27180 localabstract:scrcpy_00000001

I connect via TCP using a Node.js script, sending a properly formed 10-byte buffer for INJECT_KEYCODE (e.g., keycode 66 for ENTER).
The connection succeeds, the socket exists in /proc/net/unix, but the device never reacts to any input, and there is no logcat output indicating the control message is being handled.

Observations:

• The same control payload works fine when sent to the main scrcpy socket (@scrcpy) in a standard single-process setup
• In the control-only process (video=false), the socket is created (@scrcpy_00000001), but control events are silently ignored
• It seems that without video=true, the Device and Controller classes do not fully initialize or call reader.start() to handle messages

Question:

Is it possible to run scrcpy-server in a control-only mode with video=false, and still have control commands handled via a separate scid socket?

If not, is this a design limitation, or could it be supported in future versions?
Would setting video=true but using minimal settings (e.g., max_size=1, bit_rate=1M) be a valid workaround?

Thanks in advance for clarifying this behavior. 🙏

[rom1v]

Did you send the "dummy byte" (a single 0 on the socket), or disable it? https://github.com/Genymobile/scrcpy/blob/master/doc/develop.md#connection

[vinhpdt55]

Did you send the "dummy byte" (a single 0 on the socket), or disable it? https://github.com/Genymobile/scrcpy/blob/master/doc/develop.md#connection

Yes, I did receive and process the dummy byte.
When connecting to the scrcpy control socket (via tunnel_forward=true), my client waits for and reads the initial dummy byte (0x00) sent by the server as specified in the scrcpy connection protocol.

Here’s the relevant log:

Received dummy byte: <Buffer 00>

After receiving the dummy byte, I proceed to send control messages (e.g., INJECT_KEYCODE). This confirms that the client complies with the protocol requirement and does not skip or disable the dummy byte handshake.

Let me know if additional synchronization or delay is needed before sending the first control command.

[rom1v]

The device should receive the message sent by the client here:

scrcpy/server/src/main/java/com/genymobile/scrcpy/control/Controller.java

Line 237 in c5ed2cf

msg = controlChannel.recv();

Check that you correctly receive the data when you send your message.