speak-to-me-mcp

Give Claude Code a voice. This MCP server enables Claude to speak summaries, announcements, and explanations aloud using text-to-speech — right from your terminal.

How it works

When you use a slash command like /speak_final, Claude composes a natural, concise summary of what just happened and speaks it through your Mac's speakers. You get an audible debrief without leaving your terminal.

Two TTS engines available:

Related MCP server: TTS-MCP

Installation

Plugin Install (Recommended)

The easiest way — installs the MCP server, tools, and slash commands together:

/plugin marketplace add charlesmooredev/speak-to-me-mcp
/plugin install speak-to-me@speak-to-me

That's it — works immediately with no API keys.

Updating

/plugin marketplace update
/plugin install speak-to-me@speak-to-me

MCP-Only Install

If you just want the tools without the slash commands:

claude mcp add speak-to-me -- node "/path/to/speak-to-me-mcp/dist/index.cjs"

Building from Source

git clone https://github.com/charlesmooredev/speak-to-me-mcp.git
cd speak-to-me-mcp
npm install
npm run build
npm run bundle

(Optional) Enable OpenAI Premium Voices

For higher quality speech, add your OpenAI API key to the MCP config:

claude mcp add speak-to-me \
  -e OPENAI_API_KEY=sk-your-key-here \
  -- node "/path/to/speak-to-me-mcp/dist/index.cjs"

Or set it in your .mcp.json:

{
  "mcpServers": {
    "speak-to-me": {
      "command": "node",
      "args": ["/path/to/speak-to-me-mcp/dist/index.cjs"],
      "env": {
        "OPENAI_API_KEY": "sk-your-key-here"
      }
    }
  }
}

Slash Commands

/speak_final

Speaks a summary of the final results after Claude finishes a task. Claude will compose a natural 2-4 sentence overview of what changed — files modified, bugs fixed, features added — and read it aloud.

Example: After fixing a bug, you'll hear something like:

"I fixed the authentication timeout issue. The session handler was losing the refresh token on redirect. I updated the token storage logic and added a test to cover that edge case."

/speak_plan

Speaks Claude's current plan before implementation begins. Useful for reviewing the approach hands-free before giving the go-ahead.

Example: Before starting work, you'll hear:

"Here's my plan. I'm going to add form validation to the signup page. I'll create a validation utility, wire it up to the form inputs, and display inline error messages. The existing styles already support error states, so no CSS changes needed."

/change_voice

Lists all available voices and lets you switch to a different one. Claude will play a test phrase with the new voice so you can hear it before committing.

/speech_style

Set a speaking style or persona that changes how Claude writes spoken summaries. This doesn't change the TTS voice — it changes the personality and language Claude uses when composing text.

/speech_style talk like a clown
/speech_style speak like a business woman
/speech_style be super concise and direct
/speech_style explain things like a friendly teacher
/speech_style clear

Once set, the style applies to all spoken output (/speak_final, /speak_plan, etc.) for the session.

Examples of what you'll hear:

Use /speech_style clear to reset back to Claude's natural speaking style.

MCP Tools

These are the underlying tools that power the slash commands. Claude can also call them directly during any conversation.

speak

Speak text aloud using text-to-speech.

Audio plays in the background — the tool returns immediately without blocking Claude.

list_voices

List available TTS voices.

configure_speech

Set session defaults. Call with no arguments to view current config.

speech_style

Set a speaking style or persona. Call with no arguments to view current style.

Voice Options

System Voices (macOS)

The default voice is your macOS system voice — whatever you've set in System Settings > Accessibility > Spoken Content. This means if you have a Siri voice selected (like "American Voice 4"), that's what you'll hear.

To change your system voice: System Settings > Accessibility > Spoken Content > System Voice. Or use /change_voice to pick a specific named voice.

Other notable system voices:

Run /change_voice to hear them all and pick your favorite.

OpenAI Voices

Available when OPENAI_API_KEY is set:

Models:

• tts-1 — Fast, good quality (default)

• tts-1-hd — Slower, higher quality

Project Structure

speak-to-me-mcp/
  src/
    index.ts        Entry point
    tools.ts        MCP tool registration
    tts.ts          TTS engine logic (system + OpenAI)
    types.ts        TypeScript interfaces
    constants.ts    Defaults and limits
  skills/
    speak_final/    /speak_final slash command
    speak_plan/     /speak_plan slash command
    change_voice/   /change_voice slash command
    speech_style/   /speech_style slash command
  dist/
    index.cjs       Bundled server (run this)

Requirements

• macOS (uses say and afplay commands)

• Node.js >= 18

• OpenAI API key (optional, for premium voices)

License

MIT