Voice Mcp is an MCP server that enables bi-directional voice conversations with Claude Code and other Model Context Protocol agents. It interfaces with local speech processing tools such as Whisper.cpp and Kokoro, or cloud endpoints using an OpenAI API key for speech-to-text and text-to-speech services. Software engineers, terminal users, and multitasking developers use this tool when typing is inconvenient or hands-free operation is required, such as stepping away from the desk or debugging during manual tasks. By capturing audio input through the microphone and returning generated audio directly through system speakers, Voice Mcp allows users to speak prompts and listen to model responses. It features smart silence detection that automatically halts recording when the speaker pauses, maintaining conversational flow. Users can run the service completely offline with local machine learning models for privacy, or configure it with remote OpenAI APIs as a fallback, all managed through standard MCP server configurations.
Category: Design, Media & Creative
Tags: audio, speech, tts, voice
Follow these steps to set up Voice Mcp using the uv tool or Claude Code plugin: 1. Install platform audio dependencies, such as ffmpeg, portaudio, and audio libraries for your operating system (e.g., via brew on macOS or apt on Ubuntu). 2. If using Claude Code plugin mode, run: bash claude plugin marketplace add mbailey/voicemode claude plugin install voicemode@voicemode /voicemode:install 3. Alternatively, install via the Python package installer: bash curl -LsSf https://astral.sh/uv/install.sh | sh uvx voice-mode-install claude mcp add --scope user voicemode -- uvx --refresh --from voice-mode voicemode-mcp-launcher 4. If using cloud speech services instead of local Whisper or Kokoro, set your API key by running export OPENAI_API_KEY=your-openai-key. 5. Start conversational voice mode by executing claude converse or /voicemode:converse.
Part of MCP Servers
You can install Voice Mcp as a Claude Code plugin by adding the mbailey/voicemode marketplace and running the install command. Alternatively, install the uv package manager, execute uvx voice-mode-install to set up dependencies, and register the server into Claude Code using the claude mcp add command targeting voicemode-mcp-launcher.
Voice Mcp connects MCP-capable agents like Claude Code to speech-to-text and text-to-speech engines. It records voice input through your microphone, detects silence to stop recording automatically, sends the transcribed text to the language model, and vocalizes the model's generated response back through your speakers.
Yes, Voice Mcp supports fully local, offline voice processing. You can set up Whisper.cpp for local speech-to-text transcription and Kokoro for local text-to-speech generation. When configured locally, no external API calls or internet connections are required, keeping your audio data private on your local machine.
Voice Mcp is compatible with Linux, macOS, Windows natively or through WSL2, and NixOS. It requires Python versions 3.10 through 3.14 along with operating system audio dependencies like ffmpeg and portaudio. WSL2 users specifically require pulseaudio packages for microphone access.
Voice Mcp is open-source software licensed under the MIT license. You can view, inspect, contribute to, or fork its source code directly on GitHub at https://github.com/mbailey/voice-mcp.