# waxberry-dev/live-translate-mcp [Health: Active]

**Category:** 🌎 Translation Services  
**Repository:** https://github.com/waxberry-dev/live-translate-mcp  
**GitHub Stars:** 3  
**npm Downloads (last month):** 361  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 1  
**Directory Page:** https://allmcps.com/mcp/waxberry-dev-live-translate-mcp

## Description
Real-time English ↔ Mandarin Chinese speech translation. Transcribes audio locally with Whisper, translates via Claude API, and synthesises speech locally with Piper TTS. Pass a WAV file path and Claude handles the rest.

## Tools
Capabilities this server exposes over MCP:

- **translate_file** — Translate a WAV audio file. Pass an absolute path — the server transcribes it, translates the text via Claude, synthesises speech, saves `<name>_translated.wav` next to the original, and plays it automatically.
- **translate_speech** — Translate raw audio passed as a base64-encoded WAV string. Returns the transcription, translation, and synthesised audio as base64 WAV — useful for programmatic workflows.
- **health_check** — Verify that all dependencies (Whisper model cache, Piper voice files, `espeak-ng`) are present and ready before making a translation request.

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

```json
"mcpServers": {
  "live-translate-mcp": {
    "command": "npx",
    "args": ["-y","live-translate-mcp"],
    "env": {
      "ANTHROPIC_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `ANTHROPIC_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What waxberry-dev/live-translate-mcp MCP server does

The waxberry-dev/live-translate-mcp MCP server adds English-to-Mandarin and Mandarin-to-English speech translation tools to compatible MCP clients. It accepts WAV audio, identifies whether the speech is English or Mandarin, transcribes it with Whisper, sends the resulting text to Claude for translation, and converts the translated text back into speech with Piper TTS.

Audio processing stays on the local machine for speech recognition and synthesis. The translated text is sent to the Anthropic API; the source audio itself is not uploaded by these processing stages. The project is intended for Claude Desktop and Claude Code workflows where an agent needs to listen to a recording and return spoken output.

## How it works

For file-based requests, `translate_file` receives an absolute WAV path. The server performs the full pipeline, writes a translated WAV file beside the source using a `_translated.wav` suffix, and plays the result automatically. The tool returns the original transcription, the translated text, and the output path.

For applications that already hold audio in memory, `translate_speech` accepts a base64-encoded WAV string. Its response contains the transcription, translation, and generated speech encoded as base64 WAV data, avoiding the need to manage an input or output file through the tool.

Whisper model data and Piper voice files are downloaded on first use and cached under `~/.live-translate/` by default. The supported speech languages are English and Mandarin Chinese, with language detection handled automatically.

## Setup and configuration

Run the waxberry-dev/live-translate-mcp MCP server with Node.js 18 or newer:

```bash
npx -y live-translate-mcp
```

Set `ANTHROPIC_API_KEY` in the server environment. This credential is required because Claude performs the text translation. Install `espeak-ng` separately so Piper can phonemise text; macOS users can install it with Homebrew, while Linux users can use their distribution package manager.

Piper voice files normally use `~/.live-translate/voices`. Set `PIPER_VOICE_DIR` when the voice files are stored elsewhere. The Whisper and Piper model downloads occur automatically on the first translation request.

In Claude Desktop, register the server with the `npx` command and pass the API key through the MCP server's `env` configuration. Claude Code can add the same server through its MCP configuration, with the API key available in the process environment or MCP environment settings.

## Tools and capabilities

- `translate_file` translates an absolute-path WAV recording, saves a sibling translated WAV file, and starts playback.
- `translate_speech` handles base64-encoded WAV input and returns base64-encoded synthesized WAV output alongside the text results.
- `health_check` checks whether the Whisper cache, Piper voice files, and `espeak-ng` are available before translation.

The waxberry-dev/live-translate-mcp MCP server is suitable for meeting clips, recorded conversations, and programmatic audio workflows. It does not provide a general offline translation pipeline: Claude remains necessary for the translation step, even though transcription and speech synthesis are local.

## Limitations and notes

Input audio is documented as WAV, and file requests require an absolute path. Translation is limited to English and Mandarin Chinese. Audio processing is partly offline rather than fully offline because translated text is sent to Claude through the Anthropic API. The first request may need to download approximately 150 MB of Whisper model data and approximately 200 MB of Piper voice models, which are then cached locally.

_Full upstream README: https://allmcps.com/mcp/waxberry-dev-live-translate-mcp/readme_