MCP Audio Transcriber
A portable, Dockerized Python tool that implements a Model Context Protocol (MCP) for audio transcription using OpenAI's Whisper models—and even ships with a Streamlit-powered web UI so you can upload an audio file and download the transcription as JSON.
🚀 Features
- Modular MCP interface (
mcp.py) that defines a standard ModelContextProtocol. - Whisper-based implementation (
WhisperMCP) for high-quality, multi-language transcription. - Command-line interface (
app.py) for batch or ad-hoc transcription:python app.py <input_audio> <output_json> [--model MODEL_NAME]
- Docker support for a consistent runtime:
docker build -t mcp-transcriber .
docker run --rm \
-v /full/path/to/data:/data \
mcp-transcriber:latest \
/data/input.wav /data/output.json
- Streamlit web app (
streamlit_app.py) letting end users:- Upload any common audio file (.wav, .mp3, .ogg, .m4a)
- Choose a Whisper model size
- Preview the transcription live
- Download the JSON result with one click
📦 Prerequisites
- Python 3.10+
- ffmpeg installed & on your PATH
- (Optional) Docker Engine / Docker Desktop
- (Optional) Streamlit
🔧 Installation
- Clone the repo
git clone https://github.com/ShreyasTembhare/MCP---Audio-Transcriber.git
cd MCP---Audio-Transcriber
- Python dependencies & FFmpeg
pip install --upgrade pip
pip install -r requirements.txt
# On Ubuntu/Debian:
sudo apt update && sudo apt install ffmpeg
# On Windows:
# Download a static build from https://ffmpeg.org and add its bin/ to your PATH
- (Optional) Docker
- Install Docker Desktop
- Enable WSL integration if using WSL2.
- (Optional) Streamlit
🎯 Usage
1. CLI Transcription
python app.py <input_audio> <output_json> [--model tiny|base|small|medium|large]
<input_audio>: path to your audio file<output_json>: path where the JSON result will be saved--model: choose Whisper model size (default: base)
Example:
python app.py data/input.ogg data/output.json --model tiny
cat data/output.json
2. Docker
Build the image:
docker build -t mcp-transcriber .
Run it (mounting your data/ folder):
docker run --rm \
-v "/full/path/to/your/project/data:/data" \
mcp-transcriber:latest \
/data/input.wav /data/output.json
Then inspect:
ls data/output.json
cat data/output.json
3. Streamlit Web UI
Launch the app:
streamlit run streamlit_app.py
- Open http://localhost:8501 in your browser
- Upload an audio file
- Select the Whisper model size
- Click Transcribe
- Preview & download the resulting JSON
📁 Project Structure
MCP-Audio-Transcriber/
├── app.py # CLI entrypoint
├── mcp.py # Model Context Protocol + WhisperMCP
├── requirements.txt # Python dependencies
├── streamlit_app.py # Streamlit interface
├── Dockerfile # Container definition
├── .gitignore # ignore **pycache**, venvs, etc.
├── LICENSE # MIT license
└── data/ # sample input and output
├── input.ogg
└── output.json