Voice Mode
STDIOVoice Mode MCP server for hands-free conversation with Claude Code using speech recognition
Voice Mode MCP server for hands-free conversation with Claude Code using speech recognition
Voice Mode for Claude Code allows you to have a continuous two-way conversation with Claude Code, hands-free.
It uses the new Claude Code hooks to deliver voice input to Claude while it works.
This lets you speak continuously to Claude - interrupt, redirect, or provide feedback without stopping what Claude is doing.
Optionally enable text-to-speech to have Claude speak back to you.
Voice recognition and text-to-speech are handled by the browser, so there is nothing to download, and no API keys are needed.
Installation is easy.
npm install -g @anthropic-ai/claude-code
npx mcp-voice-hooks@latest install-hooks claude mcp add voice-hooks npx mcp-voice-hooks@latest
claude
The browser interface will automatically open after 3 seconds (http://localhost:5111).
Click "Start Listening"
Say something to Claude. You will need to send one message in the Claude Code CLI to start the conversation.
By default, utterances are sent automatically when you pause. You can switch to "Wait for Trigger Word" mode in the browser interface:
The trigger word is case-insensitive and will be automatically removed from your message before sending.
There are two options for voice responses:
Mac has built-in text to speech, but high quality voices are not available by default.
You can download high quality voices from the system voice menu: System Settings > Accessibility > Spoken Content > System Voice
Click the info icon next to the system voice dropdown. Search for "Siri" to find the highest quality voices. You'll have to trigger a download of the voice.
Once it's downloaded, you can select it in the Browser Voice (Local) menu in Chrome.
Test it with the bash command:
say "Hi, this is your Mac system voice"
To use Siri voices with voice-hooks, you need to set your system voice and select "Mac System Voice" in the voice-hooks browser interface.
Other downloaded voices will show up in the voice dropdown in the voice-hooks browser interface so you can select them there directly, instead of using the "Mac System Voice" option.
There is a bug in Safari that prevents browser text-to-speech from loading high-quality voices after browser restart. This is a Safari Web Speech API limitation. To use high-quality voices in Safari you need to set your system voice to Siri and select "Mac System Voice" in the voice-hooks browser interface.
The hooks are automatically installed/updated when the MCP server starts. However, if you need to manually install or reconfigure the hooks:
npx mcp-voice-hooks install-hooks
This will configure your project's .claude/settings.local.json with the necessary hook commands.
To completely remove MCP Voice Hooks:
# Remove from Claude MCP servers claude mcp remove voice-hooks
# Also remove hooks and settings npx mcp-voice-hooks uninstall
This will:
.claude/settings.local.jsonThe default port is 5111. To use a different port, set the MCP_VOICE_HOOKS_PORT environment variable in your project's .claude/settings.local.json:
{ "env": { "MCP_VOICE_HOOKS_PORT": "8080" } }
This environment variable is used by both:
Note: Setting this in .claude/settings.local.json is the recommended approach. The environment variable will be available to both the MCP server process and the hook commands.
When running in MCP-managed mode, the browser will automatically open if no frontend connects within 3 seconds. To disable this behavior:
{ "env": { "MCP_VOICE_HOOKS_AUTO_OPEN_BROWSER": "false" } }
Simply add the following to your project's .claude/settings.local.json and restart Claude Code:
{ "extraKnownMarketplaces": { "mcp-voice-hooks-marketplace": { "source": { "source": "git", "url": "https://github.com/johnmatthewtennant/mcp-voice-hooks.git" } } }, "enabledPlugins": { "mcp-voice-hooks-plugin@mcp-voice-hooks-marketplace": true } }
set enabled to false if you want to temporarily disable the plugin.