LocalVoice
for AI agents

Every feature can be called by an AI assistant. Everything runs locally on the Mac — the AI sends a request, LocalVoice does the work, nothing is uploaded by LocalVoice.

Connect in one click

Claude Desktop / Cursor

Open LocalVoice → Settings → AI → Connect. Restart the AI app.

Claude Code

Settings → AI → Copy Command, then paste into Terminal.

Siri & Shortcuts

Actions appear in the Shortcuts app under LocalVoice.

Manual MCP config:

{
  "mcpServers": {
    "localvoice": {
      "command": "/Applications/LocalVoice.app/Contents/MacOS/localvoice",
      "args": ["mcp"]
    }
  }
}

Tools

Tool	What it does	Changes files?
`transcribe`	Transcribe an audio or video file on this Mac. Always saves a Markdown transcript to the LocalVoice library; 'formats' also writes txt/srt/vtt/md files next to the source (or to output_dir). Use srt or vtt for subtitles. Returns the transcript text (truncated if long) and file paths. Long files can take minutes.	Yes
`list_media`	List audio/video files in a folder (default ~/Downloads), optionally only those added since a date. Use before transcribe when the user describes a recording by date ("yesterday's meeting"). Read-only.	No
`read_aloud`	Turn text, a text/Markdown file, or a PDF into a spoken M4A audio file using the Mac's built-in voices (offline). Give 'path' or 'text'. The voice is chosen automatically from the text's language unless 'voice' is set (see list_voices).	Yes
`list_voices`	List the Mac's installed speech voices, best quality first, optionally for one language (e.g. "en", "vi"). Read-only.	No
`search_transcripts`	Find where a word or phrase was said across saved transcripts (case- and accent-insensitive). Returns transcript, timestamp and the surrounding sentences. Read-only.	No
`list_transcripts`	List saved transcripts, newest first. Read-only.	No
`get_transcript`	Get the full text of a saved transcript ("latest" for the newest, an id, or part of the title), with timestamps. Use this to summarise or quote a recording. Read-only.	No
`download_model`	Download a Whisper speech model (needed for Vietnamese and other languages Apple's engine lacks). base ~150 MB (fast), small ~490 MB, turbo ~550 MB (best). Downloads from Hugging Face once; transcription itself stays offline. Ask the user before downloading.	Yes

Example requests

“Transcribe yesterday's meeting recording into Markdown”
“Make SRT subtitles for this video”
“Read this PDF aloud and save it as M4A”
“Find where we talked about 'price' in this week's transcripts”
“Get the latest transcript so you can summarize it”

Command line

CLI

localvoice list
localvoice transcribe --path ~/Downloads/meeting.m4a --formats srt

LocalVoicefor AI agents