workos-meeting-mcp-server 1.0.0

MCP server for meeting transcription & analysis — transcribe audio/video via faster-whisper, extract insights, generate notes.

Meeting MCP Server

An independent MCP (Model Context Protocol) server for meeting transcription and analysis. Transcribes audio/video files locally using faster-whisper and generates meeting insights, notes, and follow-up emails using any Ollama-compatible LLM.

Features

• 🎙️ Audio Transcription — Transcribe MP3, WAV, M4A, OGG, FLAC, AAC, WMA files using faster-whisper (local, private)
• 🎬 Video Support — Automatically extract audio from MP4, WebM, MKV, MOV, AVI files via ffmpeg
• 📊 Meeting Insights — Extract structured data: summary, key decisions, action items, sentiment
• 📝 Meeting Notes — Generate formatted markdown meeting notes
• 📧 Follow-up Email — Draft professional follow-up emails
• 🔒 100% Local — All transcription runs locally on CPU, no data leaves your machine

Installation

pip install workos-meeting-mcp-server

System Requirements

• ffmpeg must be installed for video file support:

  # macOS
  brew install ffmpeg
  # Ubuntu/Debian
  sudo apt-get install ffmpeg
  # Docker — already included in WorkOS backend image
  

Quick Start

# Basic usage (defaults to Ollama at localhost:11434)
meeting-mcp-server

# With custom Ollama URL and model
OLLAMA_URL=http://localhost:11434 \
OLLAMA_MODEL=qwen3 \
WHISPER_MODEL=small \
meeting-mcp-server

Environment Variables

Variable	Default	Description
OLLAMA_URL	http://localhost:11434	Ollama API endpoint for LLM analysis
OLLAMA_MODEL	qwen3	LLM model name for analysis tasks
WHISPER_MODEL	small	Whisper model size: tiny, base, small, medium, large-v3
UPLOAD_DIR	data/uploads	Directory to scan for uploaded files
MEETING_LOG_LEVEL	INFO	Log level

Available Tools

meeting_transcribe_audio

Transcribe an audio or video file using faster-whisper.

Parameters:

• file_path (string, required) — Path to the audio/video file
• language (string, optional) — Language code (auto-detected if empty)

Returns: Transcript text, timestamped segments, detected language, duration

meeting_extract_insights

Extract structured insights from a meeting transcript.

Parameters:

• transcript (string, required) — Full meeting transcript text

Returns: JSON with title, summary, key_decisions, action_items, topics, sentiment, attendees

meeting_generate_notes

Generate formatted meeting notes in markdown.

Parameters:

• transcript (string, required) — Meeting transcript text
• include_timestamps (boolean, default: true) — Include timestamps

Returns: Formatted markdown meeting notes

meeting_draft_follow_up

Draft a follow-up email based on meeting outcomes.

Parameters:

• summary (string, required) — Meeting summary
• action_items (string, required) — Action items (JSON or comma-separated)
• attendees (string, required) — Attendee names/emails (comma-separated)

Returns: JSON with subject and body fields

Adding to AI Applications

Claude Desktop / Cursor / Windsurf

Add to your MCP config file:

{
  "mcpServers": {
    "meeting": {
      "command": "meeting-mcp-server",
      "env": {
        "OLLAMA_URL": "http://localhost:11434",
        "OLLAMA_MODEL": "qwen3",
        "WHISPER_MODEL": "small"
      }
    }
  }
}

WorkOS

Add via Settings → MCP Servers → Add Server:

• Name: meeting
• Command: meeting-mcp-server
• Environment Variables: Set OLLAMA_URL, OLLAMA_MODEL as needed

Whisper Model Comparison

Model	Size	VRAM	Speed	Accuracy
tiny	75MB	~1GB	~32x	Good for clear audio
base	142MB	~1GB	~16x	Better accuracy
small	466MB	~2GB	~6x	Recommended
medium	1.5GB	~5GB	~2x	High accuracy
large-v3	3GB	~10GB	~1x	Best accuracy

License

MIT