# Voice Bridge [Health: Active]

**Category:** 🎧 Text-to-Speech  
**Repository:** https://github.com/Tomorrow-You/voice-bridge  
**GitHub Stars:** 0  
**npm Downloads (last month):** 138  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/voice-bridge

## Description
Multi-engine TTS for AI coding assistants. Free, 5 engines, no API key needed.

## Tools
Capabilities this server exposes over MCP:

- **speak** — Speak text aloud. Optionally override engine for this call.
- **set_engine** — Switch the default TTS engine (`auto`, `edge-tts`, `elevenlabs`, `kokoro`, `say`, `espeak`)
- **get_status** — Show current mode, engine, and available engines
- **list_voices** — List available voices. Defaults to current engine if omitted.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "voice-bridge": {
    "command": "npx",
    "args": ["-y","ai-voice-bridge"],
    "env": {
      "VOICE_BRIDGE_HOME": "",
      "ELEVENLABS_API_KEY": "",
      "VOICE_BRIDGE_MODE": "",
      "VOICE_BRIDGE_ENGINE": "",
      "VOICE_BRIDGE_EDGE_VOICE": "",
      "VOICE_BRIDGE_EDGE_RATE": "",
      "VOICE_BRIDGE_ELEVENLABS_SPEED": "",
      "VOICE_BRIDGE_KOKORO_VOICE": ""
    }
  }
}
```

**Requires environment variables:** `VOICE_BRIDGE_HOME`, `ELEVENLABS_API_KEY`, `VOICE_BRIDGE_MODE`, `VOICE_BRIDGE_ENGINE`, `VOICE_BRIDGE_EDGE_VOICE`, `VOICE_BRIDGE_EDGE_RATE`, `VOICE_BRIDGE_ELEVENLABS_SPEED`, `VOICE_BRIDGE_KOKORO_VOICE` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Voice Bridge MCP server does

Voice Bridge MCP server adds local text-to-speech controls to AI coding assistants. Its MCP interface exposes four operations: speak text aloud, change the default engine, inspect the current mode and available engines, and list voices for an engine. The package also includes command-line utilities and a Claude Code plugin, so it can be used through MCP, shell pipes, or response hooks.

Audio is played on the machine running Voice Bridge. The project is intended for desktop or workstation use with speakers or headphones, not for delivering audio over a network. A text-safety filter removes code blocks, secrets, file paths, URLs, and markdown before speech. Claude Code's hook can also limit spoken output to 2,000 characters and play it sentence by sentence.

## How it works

The package selects an engine automatically by default. The availability order is edge-tts, macOS say, espeak, Kokoro, and ElevenLabs. ElevenLabs enters that selection process only when its SDK is installed and a valid API key is configured. You can also select an engine explicitly for the default configuration or for an individual speak operation.

The available engines cover different deployment needs. edge-tts provides Microsoft neural voices without an API key. ElevenLabs is a paid cloud option. Kokoro runs locally through ONNX and requires model files. macOS say and espeak-ng use platform speech facilities and provide basic voices. Kokoro is English-only according to the project documentation.

With Claude Code, the plugin adds a Stop hook, a `/speak` skill, and the MCP server. The hook can extract content from `<speak>` tags for single-turn use, or speak the complete response in always-on mode. It runs in the background and records hook activity in a rotating log file.

## Setup and configuration

Install Python 3.10 or newer, then install the package with the engine extra you need. `ai-voice-bridge[edge]` installs the free edge-tts path, `[all]` installs all engine dependencies, and the base package supports minimal platform options. Linux or Windows also need a supported player: mpv or ffplay on Linux, and ffplay or mpv on Windows. macOS uses its built-in `afplay` utility.

Run `voice-bridge test` to verify local playback. The setup command can detect engines, test audio, optionally collect an ElevenLabs key, write initial state, and show Claude Code integration options. The Claude Code plugin can be installed from the project's marketplace, or the Stop hook can be added manually to `.claude/settings.json`.

Configuration is stored under the platform-specific Voice Bridge data directory. `VOICE_BRIDGE_HOME` overrides that location. The directory can contain `.env` for the ElevenLabs key, `.state` for mode and voice settings, `models/` for Kokoro files, and the hook log. The default mode is off, meaning speech is requested per turn; always-on mode can be enabled with `voice-bridge on`.

## Tools and capabilities

The Voice Bridge MCP server provides:

- `set_engine`: sets `auto`, edge-tts, ElevenLabs, Kokoro, say, or espeak as the default.
- `get_status`: reports the current mode, selected engine, and available engines.
- `list_voices`: lists voices, using the current engine when none is specified.

Voice discovery can be filtered by gender or locale where supported, and voices can be previewed from the command line. The CLI also supports random voice samples and engine status checks.

## Limitations and notes

Voice Bridge MCP server requires local audio hardware and a compatible player. Installation may succeed on a headless server, Docker container, SSH session, or CI runner, but speak operations will not produce audio without sound hardware and a player. ElevenLabs requires both its SDK and an API key. Kokoro requires an approximately 200 MB model download. The documentation does not identify a software license for the project.

_Full upstream README: https://allmcps.com/mcp/voice-bridge/readme_