Voice-to-Text MCP Server (vtt-mcp)

The Voice-to-Text MCP Server (vtt-mcp) enables high-accuracy, low-latency transcription of voice commands and integrates seamlessly with the Goose AI agent. By leveraging Whisper-based Speech-to-Text (STT) models, the system provides outstanding transcription while preserving user data privacy by running completely offline.

• Real-time transcription for active audio input.
• Push-to-Talk support for single-session recordings.
• Historical log management for saved transcripts.
• Linux-first design with PipeWire audio support.

The vtt-mcp requires the following tools to be installed:

1. Rust programming language. Install it using rustup.
2. PipeWire audio framework to handle system audio.
3. Whisper machine learning models downloaded from whisper.cpp.

To install and run:

• Clone the repository from GitHub.
• Navigate to the root folder and build the project using Cargo.
• Launch vtt-mcp server from the terminal.

Here are common vtt-mcp commands:

• Start real-time transcription: ./vtt-mcp --start-session
• Retrieve logs: ./vtt-mcp --get-log
• Transcribe single audio clip with specific duration: ./vtt-mcp --transcribe-clip --duration
• List all commands: ./vtt-mcp --help

Currently, there is no contribution guide (CONTRIBUTING.md). If you'd like to contribute, please open an issue or contact the maintainers.

This project is licensed under the GNU General Public License v3.0 (GPLv3). See the LICENSE file for the complete terms.