gmeet-mcp

MCP server for Google Meet — join calls, speak, listen, and chat via AI agents.

Architecture

Agent ──MCP stdio──▶ gmeet-mcp server
                        ├── SessionManager (multi-meeting)
                        ├── BrowserManager (Playwright + persistent Chrome profile)
                        ├── MeetPage (DOM automation: join, leave, chat, captions)
                        ├── AudioInjector (TTS → WAV → --use-file-for-fake-audio-capture)
                        ├── TTS Engine (OpenAI gpt-4o-mini-tts)
                        ├── AudioCapture (tab audio → WebM chunks)
                        └── Transcriber (Whisper API on captured audio)

Setup

1. Install dependencies

npm install

2. Install Playwright Chromium

npx playwright install chromium

3. Configure environment

cp .env.example .env
# Edit .env with your OpenAI API key

4. Authenticate with Google

Run the server and call the authenticate tool, which opens a headed browser for Google login. Credentials are persisted in ~/.gmeet-mcp/chrome-profile/.

Usage

As an MCP server

Add to your MCP client config:

{
  "mcpServers": {
    "gmeet": {
      "command": "npx",
      "args": ["tsx", "src/index.ts"],
      "cwd": "/path/to/gmeet-mcp",
      "env": {
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

Development

npm run dev

Tools

Tool	Description
authenticate	Open headed browser for Google login
join_meeting	Join a Google Meet by URL. Returns sessionId
leave_meeting	Leave a meeting by sessionId
speak	TTS text → audio injected into meeting mic
get_transcript	Get running transcript (speaker + text + timestamp)
send_chat_message	Send a chat message in the Meet
get_chat_messages	Read chat messages from the Meet
get_participants	List current participants
get_meeting_status	Meeting status, duration, mute state
toggle_microphone	Mute/unmute
toggle_camera	Enable/disable camera
raise_hand	Raise hand in meeting
react	Send emoji reaction
take_screenshot	Screenshot current meeting view

Configuration

All configuration can be set via environment variables or ~/.gmeet-mcp/config.json.

Variable	Default	Description
OPENAI_API_KEY	—	Required for TTS and Whisper
CHROME_EXECUTABLE_PATH	Playwright bundled	Path to Chrome/Chromium
CHROME_USER_DATA_DIR	~/.gmeet-mcp/chrome-profile	Persistent profile dir
TTS_VOICE	alloy	OpenAI TTS voice
TTS_MODEL	gpt-4o-mini-tts	OpenAI TTS model
WHISPER_MODEL	whisper-1	OpenAI Whisper model
LOG_LEVEL	info	debug, info, warn, error

How It Works

• Joining: Playwright navigates to the Meet URL, handles the pre-join screen, and clicks "Join now"
• Speaking: Text is sent to OpenAI TTS, the resulting WAV is written to a file that Chrome reads via --use-file-for-fake-audio-capture
• Listening: Captions are scraped from the Meet DOM via a MutationObserver (with optional Whisper transcription of captured audio)
• Chat: DOM automation to open the chat panel, type messages, and read incoming messages
• Auth: Uses a persistent Chrome profile so you only need to log into Google once