trnscrb

Offline meeting transcription for macOS — no cloud, no subscription.

trnscrb lives in your menu bar, listens for meetings, transcribes them locally with Whisper, and makes every transcript searchable from Claude Desktop via MCP.

---

Install

brew tap ajayrmk/tap
brew install trnscrb
trnscrb install

Or with pip / uv:

pip install trnscrb && trnscrb install
uv tool install trnscrb && trnscrb install

trnscrb install is a guided setup that handles:

• Xcode Command Line Tools (needed to build the audio capture helper)
• ScreenCaptureKit audio helper (captures meeting app audio — no virtual driver needed)
• HuggingFace token for speaker diarization (pyannote)
• Whisper small model download (~500 MB, one-time)
• Claude Desktop MCP config
• macOS permissions (Microphone, Calendar, Screen Recording)
• Launch-at-login agent

---

Quick start

trnscrb start       # launch the menu bar app

With Auto-transcribe on (the default), trnscrb detects when a meeting starts — Google Meet, Zoom, Slack Huddle, Teams, FaceTime — and begins recording automatically. When the meeting ends, it stops, transcribes, and saves.

You can also trigger manually from the menu bar: Start Transcribing / Stop Transcribing.

---

How it works

Step	What happens
Meeting detected	Mic active for 5 s + meeting app found
Recording	Audio captured via ScreenCaptureKit (meeting app) + mic
Transcription	Whisper small model, runs locally on Apple Silicon
Diarization	Speaker labels via pyannote (needs HuggingFace token)
Saved	Plain .txt in ~/meeting-notes/

---

Claude Desktop integration

After trnscrb install, Claude Desktop has these tools available:

Tool	Description
start_recording	Start capturing audio
stop_recording	Stop and transcribe in the background
recording_status	Check if recording or transcribing
get_last_transcript	Fetch the most recent transcript
list_transcripts	List all saved meetings
get_transcript	Read a specific transcript
get_calendar_context	Current or upcoming calendar event
enrich_transcript	Add summary + action items via Claude API

---

CLI

trnscrb start               # launch menu bar app
trnscrb install             # guided setup / re-check dependencies
trnscrb list                # list saved transcripts
trnscrb show <id>           # print a transcript
trnscrb enrich <id>         # summarise + action items (needs ANTHROPIC_API_KEY)
trnscrb mic-status          # live mic activity monitor — useful for debugging
trnscrb devices             # list audio input devices
trnscrb watch               # headless auto-transcribe, no menu bar

---

Anthropic API key

The enrich command and MCP enrich_transcript tool require an Anthropic API key. Add it to your shell profile:

# ~/.zshrc
export ANTHROPIC_API_KEY="sk-ant-..."

Then restart your terminal or run source ~/.zshrc. If trnscrb is already running, restart it with ./restart.sh.

---

Restarting

After code changes or environment variable updates (e.g. adding ANTHROPIC_API_KEY to .zshrc), restart trnscrb:

./restart.sh

Or manually:

pkill -f "trnscrb"
source ~/.zshrc
trnscrb start

---

Audio capture

trnscrb uses ScreenCaptureKit to capture meeting app audio directly — no virtual audio driver needed. When a meeting is detected, it captures two streams:

• ScreenCaptureKit — the meeting app's audio (remote participants), device-independent
• Microphone — your voice via the default input device

Both streams are mixed into a single recording. This works seamlessly with Bluetooth earbuds, AirPods, or any audio device — you can connect or disconnect devices mid-meeting without interrupting the recording.

Requirements:

• macOS 14 (Sonoma) or later
• Screen Recording permission (granted during trnscrb install)
• Xcode Command Line Tools (to build the audio capture helper on first install)

---

Transcript format

Meeting: Weekly Standup
Date:    2025-02-18 10:00
Duration:23:14

============================================================

[SPEAKER_00]
  00:12  Good morning, let's get started.

[SPEAKER_01]
  00:18  Morning! I finished the auth PR yesterday.

Running trnscrb enrich <id> replaces SPEAKER_00 / SPEAKER_01 with inferred names and appends a summary and action items block.

---

Requirements

• macOS 14 (Sonoma) or later
• Python 3.11+
• Apple Silicon (M1/M2/M3/M4) recommended — Whisper runs on Metal
• Xcode Command Line Tools (xcode-select --install)

---

Privacy

Everything runs on your machine. No audio or transcripts leave your device unless you explicitly run enrich, which sends the transcript text to the Claude API.

---

License

MIT