Claude Speaks — Give Claude Code a voice

The problem

I spend hours in Claude Code every day. Reading long explanations, code reviews, and debugging output gets tiring. Eyes get tired, details get missed, and the constant switch between reading and coding kills flow.

The fix

An MCP server that gives Claude a voice. 400+ neural voices, 100+ languages. One command to install. Now Claude talks while I keep my eyes on the code — like a colleague explaining things out loud.

Install

Add Claude Speaks to your Claude Code setup with a single command. Requires macOS, Node.js 18+, and Python 3 with edge-tts.

# Install the MCP server
claude mcp add claude-speaks -- npx -y claude-speaks

# Install the TTS engine (one-time)
pip install edge-tts

That's it. Claude Code can now speak. Ask it to "speak your response", or "toggle speak mode on" to have every response read aloud.

At a glance

400+ neural voices
100+ languages
1 command to install

The MCP tools

speak

Speak any text aloud with a neural voice. Customizable voice and speed.

text — what to say
voice — voice ID (optional)
rate — speed, e.g. +10% (optional)

toggle_speak

Toggle automatic speak mode. When on, Claude speaks every response without being asked.

on — auto-speak everything
off — text only

list_voices

Browse all 400+ available neural voices. Filter by language code (en, es, fr, de, etc.).

Popular voice picks

Voice ID	Description
`en-GB-RyanNeural`	British male (default) — clean, professional
`en-US-GuyNeural`	American male — casual, friendly
`en-AU-WilliamNeural`	Australian male — warm, clear
`en-US-AriaNeural`	American female — articulate, modern
`en-US-JennyNeural`	American female — warm, conversational
`en-GB-SoniaNeural`	British female — polished, articulate

Architecture

Claude Speaks is a lightweight MCP server that bridges Claude Code and Microsoft's Edge TTS neural voice engine. All audio is generated locally — no API keys, no cloud calls for synthesis.

Language — TypeScript
MCP SDK — v1.12.1, Model Context Protocol
TTS engine — Microsoft Edge TTS via edge-tts (Python wrapper)
Audio — afplay (macOS native playback)
Validation — Zod schemas
License — MIT, free and open-source

Why I built it

I wanted Claude to just tell me what it found — like a colleague talking through a problem while I keep my eyes on the code. Edge TTS gives you 400+ studio-quality neural voices for free. MCP gives you tool integration with Claude. I connected the two.

One afternoon to build. Saves me hours every week.