This functions by connecting to socket files created by mpv-sockets. The mpv script there launches mpv with unique mpv sockets at /tmp/mpvsockets/.

For each mpv socket, this attaches event handlers which tell me whenever a file in a playlist ends, whenever I seek (skip), what the current working directory/path is, and whenever I play/pause an item. Once the mpv instance quits, it saves all the events to a JSON file.

Later, I can reconstruct whether or not a file was paused/playing based on the events, how long mpv was open, and which file was playing, in addition to being able to see what file/URL I was playing.

Requires python3.8+

pip install mpv-history-daemon

For some reason I can never pinpoint, this stops working after a few days of continuous use (perhaps because of my laptop suspending?), so I wrap this with another script which restarts this every so often if there are no open mpv instances. I would recommend starting this by running:

mpv_history_daemon_restart "/your/data/dir"

Usage: mpv-history-daemon daemon [OPTIONS] SOCKET_DIR DATA_DIR

  Socket dir is the directory with mpv sockets (/tmp/mpvsockets, probably) Data dir is the
  directory to store the history JSON files

Options:
  --log-file PATH               location of logfile
  --write-period INTEGER        How often to write JSON data files while mpv is open
  -s, --scan-time INTEGER       How often to scan for new mpv sockets files in the SOCKET_DIR -
                                this is a manual scan of the directory every <n> seconds  [env
                                var: MPV_HISTORY_DAEMON_SCAN_TIME; default: 10]
  --socket-class-qualname TEXT  Fully qualified name of the class to use for socket data, e.g.,
                                'mpv_history_daemon.daemon.SocketData'. This imports the class and
                                uses it for socket data.
  --help                        Show this message and exit.

Some logs, to get an idea of what this captures:

1598956534118491075|1598957274.3349547|mpv-launched|1598957274.334953
1598956534118491075|1598957274.335344|working-directory|/home/sean/Music
1598956534118491075|1598957274.3356173|playlist-count|12
1598956534118491075|1598957274.3421223|playlist-pos|2
1598956534118491075|1598957274.342346|path|Masayoshi Takanaka/Masayoshi Takanaka - Alone (1988)/02 - Feedback's Feel.mp3
1598956534118491075|1598957274.3425295|media-title|Feedback's Feel
1598956534118491075|1598957274.3427346|metadata|{'title': "Feedback's Feel", 'album': 'Alone', 'genre': 'Jazz', 'album_artist': '高中正義', 'track': '02/8', 'disc': '1/1', 'artist': '高中正義', 'date': '1981'}
1598956534118491075|1598957274.342985|duration|351.033469
1598956534118491075|1598957274.343794|resumed|{'percent-pos': 66.85633}
1598956534118491075|1598957321.3952177|eof|None
1598956534118491075|1598957321.3955588|mpv-quit|1598957321.395554
Ignoring error: [Errno 32] Broken pipe
Connected refused for socket at /tmp/mpvsockets/1598956534118491075, removing dead socket file...
/tmp/mpvsockets/1598956534118491075: writing to file...

More events would keep getting logged, as I pause/play, or the file ends and a new file starts. The key for each JSON value is the epoch time, so everything is timestamped.

By default, this scans the socket directory every 10 seconds.

This does not come with a built-in inotify/directory watcher, but it does allow you to send a signal (in particular, RTMIN) to the daemon process to check if new files have been added.

So, I have a script like this (say, mpv_signal_daemon) which sends the signal:

pkill -f 'python3 -m mpv_history_daemon daemon' -RTMIN || true

And then I run watchfiles in the background like:

watchfiles mpv_signal_daemon '/tmp/mpvsockets/'

Whenever watchfiles sees a file added/modified/deleted, it sends a signal to the daemon, to recheck if there are new sockets to process.

Note: you don't have to use watchfiles, you're free to send the signal the daemon in whatever way works for you.

I personally run this with --scan-time 30 and watchfiles. watchfiles will typically pick up all changes, but the poll is there just in case it fails or misses something

You can pass a custom socket data class with to daemon with --socket-class-qualname, which lets you customize the behaviour of the SocketData class. For example, I override particular events (see SocketDataServer) to intercept data and send it to my currently_listening server, which among other things displays my currently playing mpv song in discord:

The daemon saves the raw event data above in JSON files, which can then be parsed into individual instances of media:

Usage: mpv-history-daemon parse [OPTIONS] DATA_FILES...

  Takes the data directory and parses events into Media

Options:
  --all-events  return all events, even ones which by context you probably
                didn't listen to
  --debug       Increase log verbosity/print warnings while parsing JSON files
  --help        Show this message and exit.

As an example:

{
  "path": "/home/data/media/music/MF DOOM/Madvillain - Madvillainy/04 - Madvillain - Bistro.mp3",
  "is_stream": false,
  "start_time": 1614905952,
  "end_time": 1614906040,
  "pause_duration": 20.578377723693848,
  "media_duration": 67.578776,
  "media_title": "04 - Madvillain - Bistro.mp3",
  "percents": [
    [1614905960, 11.150022],
    [1614905981, 11.151141]
  ],
  "metadata": {}
}

This can also be called from python:

>>> from pathlib import Path
>>> from mpv_history_daemon.events import history
>>> list(history([Path("1611383220380934268.json")]))
[
  Media(path='/home/data/media/music/MF DOOM/Madvillain - Madvillainy/05 - Madvillain - Raid [feat. M.E.D. aka Medaphoar].mp3',
  is_stream=False,
  start_time=datetime.datetime(2021, 1, 23, 6, 27, tzinfo=datetime.timezone.utc),
  end_time=datetime.datetime(2021, 1, 23, 6, 29, 30, tzinfo=datetime.timezone.utc),
  pause_duration=0.0,
  media_duration=150.569796,
  media_title='Raid [feat. M.E.D. aka Medaphoar]',
  percents=[(datetime.datetime(2021, 1, 23, 6, 27, 2, tzinfo=datetime.timezone.utc), 1.471624)]
  metadata={})
]

After a while using this, I end up with thousands of JSON files in my data directory, which does use up some unnecessary space, and increases time to parse since it has to open thousands of files.

Those can be merged into a single file (which parse can still read fine) using the merge command:

Usage: mpv-history-daemon merge [OPTIONS] DATA_FILES...

  merges multiple files into a single merged event file

Options:
  --move DIRECTORY         Directory to move 'consumed' event files to, i.e.,
                           a 'remove' these from the source directory once
                           they've been merged
  --write-to PATH          File to merge all data into  [required]
  --mtime-seconds INTEGER  If files have been modified in this amount of time,
                           don't merge them
  --help                   Show this message and exit.

Merged files look like:

{ "mapping": {
      "1611383220380934268.json":
        { "1619915695.2387643": { "socket-added": 1619915695.238762 } },
      # other stuff...
    } }

... saving the filename and the corresponding data from the original files

It doesn't merge any event files who've recently (within an hour) been written to, to avoid possibly interfering with current files the daemon may be writing to.

If you want to automatically remove files which get merged into the one file, you can use the --move flag, like:

mpv-history-daemon merge ~/data/mpv --move ~/.cache/mpv_removed --write-to ~/data/mpv/"merged-$(date +%s).json"

That takes any eligible files in ~/data/mpv (merged or new event files), merges them all into ~/data/mpv/merged-...json (unique filename using the date), and then moves all the files that were merged to ~/.cache/mpv_removed (moving them to some temporary directory so you can review the merged file, instead of deleting)

My personal script which does this is synced up here