ClaudeVoice

ClaudeVoice is an open-source, MIT-licensed tool by sunlinhua-dotcom that makes Claude Code sessions audible. When a project finishes or needs attention, it speaks which project and what happened, and can optionally send an alert to a phone. The latest listed release is v2.1.0.

What It Is

ClaudeVoice installs global Claude Code hooks under ~/.claude, so every project and newly opened session is covered. A small local Node server, which uses only Node built-ins, receives events from hook.js, broadcasts them over SSE and synthesizes speech. A web panel or the macOS menubar app plays the audio. The hook exits quickly if ClaudeVoice is offline, so Claude Code is not blocked.

How the Alerts Work

Hooks fire on Stop, Notification and SubagentStop events. Speech defaults to Microsoft Edge neural voices, which need no API key, with optional Azure as a fallback. Each project gets a stable color and one of four voices (茉莉, 冰糖, 白桦, 苏打), and projects.json can pin names, colors and voices.

Phone Notifications and Remote Control

The local server can push Feishu cards directly after a setup wizard resolves the user's Feishu account. An optional Cloudflare Worker adds a public panel, PWA/Web Push, Telegram and away-from-Mac access. A separate bridge component (cc-bridge) lets power users drive Claude Code from a Feishu chat with voice or text tasks, approval cards and file transfer.

Setup Path

A one-line curl installer checks Git and Node 14+, downloads the project, runs the setup wizard and can install the prebuilt menubar app. The app can also be downloaded from GitHub Releases or built from source with swiftc on macOS 13+.

Tradeoffs to Know

The README says the project is macOS-first, with a Chinese-first voice experience. A player (web tab or menubar app) must be active to produce audio. Edge TTS uses Microsoft's public read-aloud endpoint without an SLA. The release app is ad-hoc signed rather than notarized, so first launch may require right-click then Open.