transcriber-mcp

# Transcriber MCP Transcriber MCP is a server compliant with the Model Context Protocol (MCP) that converts audio and video files into text. It utilizes faster-whisper to provide a lightweight and practical transcription server that operates in a CPU environment. ## Features - Implementation of a server compliant with the MCP protocol - Accepts audio and video files (mp3, mp4, wav, mov, avi) and performs transcription - Outputs results in text file format - Provides a communication interface with MCP clients ## Installation ### Prerequisites - Python 3.8 or higher - faster-whisper - ffmpeg (for audio and video file processing) ### Installation Steps 1. Clone the repository ```bash git clone https://github.com/yourusername/transcriber-mcp.git cd transcriber-mcp ``` 2. Create a virtual environment and install dependencies ```bash # Create a virtual environment using uv uv venv # Install dependencies uv pip install -r requirements.txt ``` ## Usage ### Starting the Server ```bash # Start the server using uv uv run -m src.main ``` ### Verify Functionality Using Client Example ```bash # Create a test audio file uv pip install gtts uv run -c "from gtts import gTTS; tts = gTTS('This is a test audio file. We will check if the transcription works correctly.', lang='en'); tts.save('test_audio.mp3')" # Execute transcription uv run -m src.client_example test_audio.mp3 ``` ### Usage from MCP Client Send a request from a client compliant with the MCP protocol as follows: ```json { "jsonrpc": "2.0", "id": 1, "method": "transcribe", "params": { "file_path": "/path/to/your/audio_or_video_file.mp3" } } ``` ### Example Response ```json { "jsonrpc": "2.0", "id": 1, "result": { "result": "/path/to/output/audio_or_video_file_transcribed.txt" } } ``` ## Configuration with Cline By using Cline in conjunction, you can perform transcription through interaction with LLM. ### Cline Configuration Add the following settings to the Cline configuration file (usually located at `~/.config/cline/settings/mcp_settings.json`): ```json { "transcribe": { "command": "uv", "args": [ "run", "--directory", "/path/to/transcriber-mcp", "python", "-m", "src.main", "--model-size=base" ] } } ``` * Please adjust the path for `--directory` according to your actual environment. * The `--model-size` can be selected from "tiny", "base", "small", "medium", "large". ## Supported File Formats - Audio files: mp3, wav - Video files: mp4, mov, avi ## Changing Model Size You can change the model size to improve transcription accuracy. ```python # Edit src/transcriber.py self.model_size = "medium" # Choose from tiny, base, small, medium, large ``` Using a larger model improves accuracy but increases memory usage and loading time. ## Future Expansion Plans - Timestamped transcription - Multilingual support - Model switching functionality ## License This project is licensed under the MIT License. Please refer to the LICENSE file for details.