Meetey

Join a meeting, run /meetey start, and get back dated notes where every decision and action item carries the [mm:ss] it came from. Audio comes off the meeting app through macOS ScreenCaptureKit, whisper.cpp does the transcription, and Claude writes the summary. Screen content — slides, screen shares, code — is opt-in per meeting, with text recognized on-device.

Recording and transcription happen entirely on your machine. No subscriptions, no third-party meeting bots, no cloud transcription service. See What stays local for exactly what is and isn't sent anywhere.

Install

Requires macOS 13+, Node.js 18+, and Xcode Command Line Tools.

npx jej2k5/meetey install

This will:

• Install whisper-cpp via Homebrew (if not already present)

• Download the Whisper base model (~142 MB) to ~/.meetey/models/

• Build and sign the Swift capture binary

• Register the MCP server in Claude Code

• Install the /meetey skill

• Add hotkeys (Ctrl+Shift+R / Ctrl+Shift+S)

After installation, open System Settings → Privacy & Security → Screen Recording and enable your terminal app. Then start a new Claude Code session.

Related MCP server: Voice MCP

Usage

In Claude Code

/meetey start    Detect running meeting apps and start recording
/meetey stop     Stop recording, transcribe, and save notes
/meetey status   Check whether a recording is active

/meetey list     Browse past meetings
/meetey show     Open one meeting's notes
/meetey search   Find a moment across every meeting
/meetey delete   Remove a recording
/meetey doctor   Check the install

/meetey start asks whether to also capture screen content. Answer up front to skip the question:

/meetey start --video       Audio + screen keyframes
/meetey start --audio-only  Audio only

Or use the hotkeys:

Workflow

1. Join your meeting in Chrome (Google Meet / Teams web), Zoom, or Microsoft Teams

2. Run /meetey start — Claude detects the running app, asks whether to capture screen content, and begins recording

3. When the meeting ends, run /meetey stop — Claude transcribes the audio, reads any screen content, and produces:

## Q3 Roadmap and API Cutover

**Recorded:** Fri 1 Aug 2026, 2:02–2:47 PM · 45 min

### Summary
...

### Key Decisions
- **[12:04]** Ship v1.1 without the motion heuristic; revisit next cycle

### Action Items
- [ ] **[31:17]** Draft the migration doc — Priya · due Friday

### Screen Content          ← only when screen capture was used
- **[04:12]** Burndown chart, actual tracking ~11 days behind plan

### Transcript
Full transcript: `2026-08-01-1402-q3-roadmap-transcript.md` · 6,180 words

Every claim carries the timestamp it came from. If you doubt a decision, jump straight to that moment in the transcript instead of searching thousands of words.

Notes are written to ~/.meetey/recordings/ as YYYY-MM-DD-HHMM-<slug>.md — chronologically sortable, so ls is a meeting history. The transcript lives in a sibling -transcript.md file rather than inline, so the notes stay scannable.

When the transcription is unreliable, the notes say so at the top instead of summarizing noise with a confident voice:

> ⚠️ Transcription quality: poor — 80% of segments were silence markers,
> fragments, or repeats. Summary may be incomplete; ggml-base.en.bin was used.

That assessment is computed from the transcript itself (fragment rate, whisper looping, speech pace against spoken time), not guessed.

Supported apps

Chrome note: ScreenCaptureKit captures at the process level, not the tab level. Mute any other Chrome tabs playing audio before starting — and if you opt into screen capture, close or hide anything you don't want recorded, because every Chrome tab and window is in frame.

Screen capture

Meetey can also capture what was on screen — slides, screen shares, code, diagrams — and fold it into the notes.

It is off by default and opt-in per meeting. Claude asks every time you run /meetey start; there is no setting that turns it on permanently. Skip the question with /meetey start --video or /meetey start --audio-only.

/meetey start --video     Record audio + screen keyframes
/meetey start --audio-only  Record audio only, don't ask

Instead of recording video, Meetey samples the screen once a second and keeps a JPEG only when the picture materially changes. An hour-long meeting typically yields a few dozen keyframes rather than 3,600, which is what makes this cheap enough to be worth doing:

Text is recognized on-device with the macOS Vision framework, so the default path sends only text. Claude reads an actual keyframe image only when the text isn't enough — a diagram, a chart, a UI screenshot.

⚠️ Screen capture records everything the target app displays — other tabs, other windows of that app, and notification banners that appear over it. A screen share can contain credentials, customer data, or a colleague's private dashboard. Close or mute anything sensitive before you opt in.

Keyframes are written to ~/.meetey/recordings/<session>-frames/ and are not deleted automatically — the notes reference them, so removing them would make the summary unverifiable. Delete that folder yourself when you're done with it.

Tuning, if you need it (pass through to meetey-capture):

Camera-heavy calls are a poor fit. A moving video tile changes the picture constantly and will burn through the frame budget. Screen capture pays off for slides, demos, and screen shares; for a faces-only discussion, stay on audio.

Managing your meetings

Everything about the library is reachable from Claude Code — there's no separate app, dashboard, or web UI to open. Every subcommand has a plain-language equivalent, so use whichever you'd reach for:

/meetey status is about the recording running right now; /meetey doctor is about whether the install itself is healthy.

Search is the one worth knowing about. Asking "what did we decide about the API cutover?" returns the decision, the meeting, and the timestamp — so you can jump straight to that second in the transcript instead of reading it.

What stays local

Nothing is uploaded to a third-party meeting service, and no bot joins your call. Audio and images never leave your machine unless Claude needs a specific keyframe to interpret a diagram. Summarization itself runs through Claude, so the text does go to the API — the same as anything else you'd paste into Claude Code.

CLI

npx jej2k5/meetey install   # First-time setup
npx jej2k5/meetey update    # Rebuild binary and refresh MCP server + skill
npx jej2k5/meetey status    # Show what's installed and whether everything is wired up

status output:

Meetey status

  ✔  meetey-capture binary  (~/.meetey/meetey-capture/.build/release/meetey-capture)
  ✔  binary code-signed
  ✔  whisper-cli  (/opt/homebrew/bin/whisper-cli)
  ✔  Whisper model (ggml-base.en.bin)  (~/.meetey/models/ggml-base.en.bin)
  ✔  MCP server files  (~/.meetey/mcp-server)
  ✔  MCP server registered in ~/.claude.json
  ✔  /meetey skill installed
  ✔  Hotkeys registered  (Ctrl+Shift+R / Ctrl+Shift+S)

Requirements

• macOS 13 (Ventura) or later

• Node.js 18+

• Xcode Command Line Tools (xcode-select --install)

• Claude Code

• Homebrew (installed automatically if missing)

Whisper model

The default model is ggml-base.en.bin (~142 MB), stored at ~/.meetey/models/. It handles clear English speech well. For better accuracy on accented speech or technical vocabulary, switch to a larger model:

To switch, download the model to ~/.meetey/models/ and set MEETEY_MODEL in the mcpServers.meetey.env section of ~/.claude.json.

Troubleshooting

How it works

/meetey skill → MCP server (Node.js, ~/.meetey/mcp-server/) → meetey-capture (Swift, ~/.meetey/)
                                                             → whisper-cli

meetey-capture is a Swift CLI using ScreenCaptureKit. It takes --app <bundle-id> and --output <path.wav>, records app audio as 16-bit PCM WAV at 16 kHz mono, and stops on SIGTERM. With --video it also samples the screen, keeps a JPEG when the picture changes, and runs Vision OCR on each one.

MCP server manages the capture process lifecycle and shells out to whisper-cli for transcription. Registered globally in ~/.claude.json so it's available in every Claude Code session.

/meetey skill drives the user-facing flow: list_apps → start_recording → stop_recording → transcribe → get_keyframes, then formats and saves the output.

Run meetey-capture --selftest to exercise the keyframe pipeline (scene detection, JPEG encoding, OCR, manifest) with synthetic frames — no Screen Recording permission or live meeting required.

Files written

Contributing

Issues and pull requests are welcome. To work on Meetey from a local clone:

git clone https://github.com/jej2k5/meetey.git
cd meetey
node cli/index.js install     # build, sign, and register everything locally

After changing Swift sources, rebuild and re-sign:

cd meetey-capture
swift build -c release
codesign --force --sign - --options runtime --entitlements entitlements.plist \
  .build/release/meetey-capture

Then restart Claude Code so the MCP server reconnects.

New source files should carry the Apache 2.0 header used by the existing files. By contributing, you agree that your contributions are licensed under the Apache License 2.0.

License

Copyright 2026 John Joseph

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0.

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the LICENSE file for the specific language governing permissions and limitations under the License.

Third-party attributions are listed in NOTICE.