snapcaster #

Turn your Mac's audio output into a Snapcast or Squeezebox stream.

snapcaster installs a virtual audio output device named Snapcaster on macOS. Pick it in System Settings → Sound like any other speaker, and everything your Mac plays — Spotify, YouTube, system sounds, all of it — is forwarded as raw PCM to:

• a Snapcast server (snapserver), for synchronized multiroom audio on any snapclient,
• a SlimProto server such as squeezed, which serves any Squeezelite / Squeezebox player,
• or any TCP socket, unix socket, or FIFO of your own.

macOS only. snapcaster is built on the Core Audio HAL and supports nothing else, on purpose. On Linux, snapserver can read from ALSA/PipeWire directly.

How it works #

                 macOS apps (Spotify, browser, system sounds…)
                                   │  play to selected output
                                   ▼
                  ┌─────────────────────────────────┐
                  │  "Snapcaster" virtual device    │   HAL driver (driver/)
                  │  output stream ──► ring buffer  │   inside coreaudiod
                  │  input  stream ◄── (loopback)   │
                  └─────────────────────────────────┘
                                   │  captured with Core Audio
                                   ▼
                  ┌─────────────────────────────────┐
                  │  snapcaster CLI (this crate)    │   user space
                  │  f32 → s16le/s32le conversion   │
                  └─────────────────────────────────┘
                                   │  raw PCM bytes
                     ┌─────────────┼─────────────────┐
                     ▼             ▼                 ▼
              tcp://host:port  unix:///path    pipe:///path
              (snapserver,     (squeezed)      (/tmp/snapfifo)
               squeezed)

Two pieces, and the split is forced by macOS, not by taste:

1. driver/ — a Core Audio HAL AudioServerPlugIn (Rust cdylib packaged as SnapcasterAudio.driver). It publishes the selectable output device and loops whatever is played into a hidden input stream. It runs inside coreaudiod, whose sandbox forbids network access — so the driver cannot talk to your server itself.
2. snapcaster CLI — a normal user-space process that opens the loopback input, converts the audio, and does the networking, with automatic reconnection.

Installation #

Homebrew #

brew install tsirysndr/tap/snapcaster
sudo snapcaster-install-driver   # installs the HAL driver into /Library, restarts coreaudiod

brew install puts the snapcaster CLI on your PATH and pulls in squeezed as a dependency. The audio driver lives in /Library and needs sudo, so it's a separate one-time step — snapcaster-install-driver ad-hoc code-signs the bundle, installs it, and restarts coreaudiod. Remove it later with sudo snapcaster-uninstall-driver.

One-line install #

curl -fsSL https://raw.githubusercontent.com/tsirysndr/snapcaster/main/install.sh | bash

This downloads the latest universal build from GitHub Releases, ad-hoc code-signs the driver, installs it plus the snapcaster CLI, and restarts coreaudiod (it will prompt for sudo). Pin a version with SNAPCASTER_VERSION=v0.1.0 in front of the command. macOS only — the script refuses to run on anything else.

From a release tarball #

Download snapcaster-<version>-macos-universal.tar.gz from the releases page, then:

tar -xzf snapcaster-*-macos-universal.tar.gz
cd snapcaster-*-macos-universal   # or wherever it extracted
./install.sh                      # installs driver + CLI, restarts coreaudiod (sudo)

The binaries are universal (Apple Silicon + Intel). Uninstall with ./uninstall.sh.

From source #

Requirements: macOS, Rust (stable), sudo for the driver install.

git clone https://github.com/tsirysndr/snapcaster
cd snapcaster

# 1. Build + install the virtual audio driver, restart coreaudiod (sudo)
./scripts/install-driver.sh

# 2. Install the CLI
cargo install --path .

After step 1, Snapcaster appears as an output device in System Settings → Sound and in the volume menu (⌥-click the menu bar icon). The driver is ad-hoc signed, which is fine for a locally built install.

The first time the CLI runs, macOS shows a microphone permission prompt — that is the capture side of the loopback device, not an actual microphone. Allow it or you'll get silence.

Quick start #

Snapcast server on another machine (e.g. 192.168.1.10):

# on the server
snapserver   # with the [stream] config below

# on the Mac
snapcaster --sink tcp://192.168.1.10:4953

Then select Snapcaster as the output device and press play on anything.

Usage #

snapcaster [OPTIONS] [COMMAND]

Commands:
  devices   List audio devices visible to Core Audio

Options:
  -s, --sink <URI>         Sink for the PCM stream         [default: tcp://127.0.0.1:4953]
  -d, --device <NAME>      Capture device name             [default: Snapcaster]
  -b, --bits <BITS>        PCM depth, 16 or 32 (LE signed) [default: 16]
  -r, --sample-rate <HZ>   44100, 48000, 88200 or 96000    [default: 44100]

Sink URIs #

URI	Meaning
tcp://host:port	Connect to a listening TCP server (snapserver tcp source, squeezed tcp input)
unix:///path/to.sock	Connect to a listening unix socket (squeezed unix input)
pipe:///path/to/fifo	Open a FIFO for writing (snapserver pipe source)

The stream is always raw PCM, interleaved, little-endian signed, 2 channels. --sample-rate switches the virtual device to the requested rate (default 44100 Hz), so no trip to Audio MIDI Setup is needed. snapcaster logs the exact format at startup; the receiving server must be configured to match.

Examples #

# Snapcast, local server, defaults (tcp://127.0.0.1:4953, 44100:16:2)
snapcaster

# 48 kHz instead of the 44.1 kHz default
snapcaster --sample-rate 48000 --sink tcp://192.168.1.10:4953

# Snapcast on the network
snapcaster --sink tcp://192.168.1.10:4953

# Snapcast on the same Mac via FIFO
snapcaster --sink pipe:///tmp/snapfifo

# squeezed / Squeezebox players
snapcaster --sink tcp://192.168.1.10:4711
snapcaster --sink unix:///tmp/squeezed.sock

# 32-bit PCM instead of 16
snapcaster --bits 32 --sink tcp://192.168.1.10:4953

# verbose logging
SNAPCASTER_LOG=debug snapcaster

Server configuration #

Snapcast (snapserver) #

sampleformat must match what snapcaster logs at startup:

# /etc/snapserver.conf
[stream]
# snapcaster connects to this port:
source = tcp://0.0.0.0:4953?name=Snapcaster&mode=server&sampleformat=44100:16:2

# — or, same machine via FIFO —
source = pipe:///tmp/snapfifo?name=Snapcaster&sampleformat=44100:16:2

squeezed (Squeezebox / Squeezelite) #

squeezed --source tcp --tcp-bind 0.0.0.0:4711 \
         --sample-rate 44100 --bits 16 --channels 2

# — or via unix socket —
squeezed --source unix --path /tmp/squeezed.sock \
         --sample-rate 44100 --bits 16 --channels 2

Run at login (launchd) #

A LaunchAgent example ships in dist/com.tsirysndr.snapcaster.plist. Edit its --sink argument for your server, then:

cp dist/com.tsirysndr.snapcaster.plist ~/Library/LaunchAgents/
launchctl bootstrap gui/$(id -u) ~/Library/LaunchAgents/com.tsirysndr.snapcaster.plist

It must be a per-user LaunchAgent, not a system LaunchDaemon: audio capture needs your login session and the microphone permission granted to your user. Logs go to /tmp/snapcaster.log; stop it with:

launchctl bootout gui/$(id -u)/com.tsirysndr.snapcaster

Audio format details #

• The virtual device runs 2 channels of 32-bit float internally.
• Supported sample rates: 44100 (default), 48000, 88200, 96000 Hz — set with --sample-rate, which switches the device itself; remember to keep the server's sampleformat in sync.
• --bits converts to 16-bit (default) or 32-bit signed little-endian on the way out.
• The macOS volume slider, volume keys and mute work on the Snapcaster device and scale the audio sent to the server (applied in the driver with a perceptual s³ curve, -96..0 dB).
• Latency: the driver adds none of its own (zero latency / zero safety offset); end-to-end delay is dominated by the server's buffering (snapserver defaults to ~1 s for multiroom sync).

Troubleshooting #

"Snapcaster" doesn't appear in Sound settings. Re-run ./scripts/install-driver.sh and check the system log: open Console.app and search for Snapcaster driver (the driver logs through syslog) or run log show --last 5m --predicate 'process == "coreaudiod"'.

The CLI says the capture device is not found. The driver isn't installed, or coreaudiod wasn't restarted. Run snapcaster devices — the input list must contain Snapcaster.

Connected, but the server plays silence. Snapcaster must be the selected output device, and the CLI needs microphone permission (System Settings → Privacy & Security → Microphone — your terminal must be allowed).

Audio is distorted / plays at the wrong speed. Sample-format mismatch. Make the server's sampleformat (or squeezed's --sample-rate/--bits/--channels) match exactly what snapcaster logs at startup.

Sink down? snapcaster keeps capturing, drops the audio, and retries the connection every 2 s — it logs each reconnect.

Uninstall #

./scripts/uninstall-driver.sh                                  # driver + coreaudiod restart
launchctl bootout gui/$(id -u)/com.tsirysndr.snapcaster 2>/dev/null || true
cargo uninstall snapcaster

Development #

cargo build                              # CLI
cargo build --release -p snapcaster-driver   # HAL driver cdylib
./scripts/install-driver.sh              # assemble bundle + install + restart coreaudiod

Workspace layout:

src/            CLI: cli.rs (clap), capture.rs (cpal), sink.rs (tcp/unix/pipe writer)
driver/         HAL AudioServerPlugIn cdylib + Info.plist (CFPlugIn factory)
scripts/        install-driver.sh / uninstall-driver.sh
dist/           launchd LaunchAgent example

The driver is modeled on Apple's NullAudio sample / BlackHole: one device object, one output and one input stream sharing a 16384-frame ring buffer, zero-timestamps derived from mach_absolute_time, and sample-rate changes via the host's RequestDeviceConfigurationChange.

One non-obvious detail: on current macOS the HAL delivers the mixed system output through the ProcessOutput and 'rite' IO operations, not WriteMix. The driver must return true from WillDoIOOperation for those ops or macOS silently refuses to start the output IOProc (output plays to nowhere and capture reads only silence). We capture the post-mix audio from WriteMix/'rite' and merely accept ProcessOutput (it is per-stream pre-mix, so writing it to the ring would drop other apps' audio when several play at once).

License #

MIT

CI pipelines #

Portable GitHub Actions jobs use PureScript Dagger modules. See that guide for commands, caching, and the jobs that remain native.