talkMCP — Voice for Claude Code

Claude Code speaks to you using AI-generated voice. When Claude has updates, completions, or needs your attention, it calls the talk tool and you hear it spoken aloud.

How It Works

1. Remote MCP server (Firebase Cloud Function) — Claude Code calls the talk tool
2. OpenAI Realtime API generates voice audio from the text
3. Firebase RTDB pushes a playback event to your machine
4. Local daemon picks up the event and plays the audio

Quick Start

1. Install the local audio daemon

npx @che-systems/talkmcp install

This sets up a macOS LaunchAgent that runs in the background and auto-plays voice messages.

Linux users: Run the listener manually (or add to systemd):

npx @che-systems/talkmcp listen

2. Add the MCP to Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "talkmcp": {
      "url": "https://us-central1-talkmcp-9dca1.cloudfunctions.net/mcp"
    }
  }
}

3. Done

Claude Code now has access to the talk and configure_voice tools. Ask Claude to "say hello" to test it.

Audio Playback Prerequisites

| Platform | Player | Install | |----------|--------|---------| | macOS | afplay | Pre-installed, no action needed | | Linux (PulseAudio) | paplay | sudo apt install pulseaudio-utils | | Linux (ALSA) | aplay | sudo apt install alsa-utils | | Windows | PowerShell SoundPlayer | Pre-installed, no action needed |

Voice Presets

| Preset | Style | Description | |--------|-------|-------------| | default / fast | Speed reader | Fast, direct, 1-2 sentences, no filler | | nosebuddy | ASMR whisper | Soft, slow, warm, whispery |

Available Voices

ash, ballad, coral, sage, verse

Tools

talk

Speaks a message aloud to the user.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | message | string | Yes | The text to speak | | preset | string | No | Voice preset (default, fast, nosebuddy) | | voice | string | No | OpenAI voice name | | systemPrompt | string | No | Custom system prompt override |

configure_voice

View available voice settings and presets.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | preset | string | No | Preset to inspect | | voice | string | No | Voice to set | | systemPrompt | string | No | Custom system prompt |

Managing the Daemon

# Install (macOS LaunchAgent)
npx @che-systems/talkmcp install

# Run directly
npx @che-systems/talkmcp listen

# Uninstall
npx @che-systems/talkmcp uninstall

# Check logs
tail -f /tmp/talkmcp.out
tail -f /tmp/talkmcp.err

Self-Hosting

To deploy your own instance:

1. Clone this repo
2. Set up Firebase: firebase login && firebase use <your-project>
3. Set your OpenAI key: create functions/.env with OPENAI_API_KEY=sk-...
4. Deploy: cd functions && npm install && npm run deploy
5. Update the MCP URL in .mcp.json to your function URL