Content
# Claude Speaks
Give Claude Code a voice. An MCP server that lets Claude speak to you using natural-sounding neural voices.
Instead of reading long responses, just ask Claude to say it out loud.
## How it works
Claude Speaks adds a `speak` tool to Claude Code. When Claude calls it, you hear the response through your speakers using Microsoft's neural TTS voices — the same ones used in Edge browser's "Read Aloud" feature.
## Quick Start
### 1. Install edge-tts (one-time)
```bash
pip install edge-tts
```
### 2. Add to Claude Code
```bash
claude mcp add claude-speaks -- npx -y claude-speaks
```
### 3. Use it
Start a new Claude Code session and ask Claude to speak:
- "Say hello to me"
- "Read your last response out loud"
- "Explain this code and speak it to me"
- "Speak this in a British accent"
## Available Tools
### `speak`
Speaks text aloud through your speakers.
| Parameter | Required | Description |
|-----------|----------|-------------|
| `text` | Yes | The text to speak |
| `voice` | No | Voice ID (default: `en-GB-RyanNeural`) |
| `rate` | No | Speed adjustment like `+10%` or `-20%` |
### `list_voices`
Lists all available neural voices.
| Parameter | Required | Description |
|-----------|----------|-------------|
| `language` | No | Filter by language code (e.g., `en`, `es`, `fr`) |
## Voices
Some popular voices:
| Voice ID | Description |
|----------|-------------|
| `en-GB-RyanNeural` | British male (default) |
| `en-US-GuyNeural` | American male |
| `en-AU-WilliamNeural` | Australian male |
| `en-US-AriaNeural` | American female |
| `en-US-JennyNeural` | American female, warm |
| `en-GB-SoniaNeural` | British female |
Run `list_voices` with `language: "en"` to see all English voices, or omit the filter to see all 400+ voices across 100+ languages.
## Requirements
- **macOS** (uses `afplay` for audio playback)
- **Node.js 18+**
- **Python 3** with `edge-tts` installed (`pip install edge-tts`)
- **Claude Code** CLI
## How to install edge-tts
```bash
# Using pip
pip install edge-tts
# Or using pipx (isolated install)
pipx install edge-tts
# Verify it works
edge-tts --text "Hello" --write-media /tmp/test.mp3 && afplay /tmp/test.mp3
```
## Manual Installation
If you prefer to install manually instead of using `npx`:
```bash
# Clone and build
git clone https://github.com/rajvardhman/claude-speaks.git
cd claude-speaks
npm install
npm run build
# Add to Claude Code
claude mcp add claude-speaks -- node /path/to/claude-speaks/dist/index.js
```
## Troubleshooting
**"Failed to speak: edge-tts not found"**
- Install edge-tts: `pip install edge-tts`
- Make sure it's in your PATH: `which edge-tts`
**No audio output**
- Check your system volume
- Test directly: `edge-tts --text "test" --write-media /tmp/test.mp3 && afplay /tmp/test.mp3`
**Voice sounds robotic**
- Don't pass the `rate` parameter — omitting it gives the most natural sound
- Try different voices: ask Claude to `list_voices` with `language: "en"`
## License
MIT
