Claude Code text to speech
Claude Code doesn’t read replies aloud by itself. You can add it with a Stop hook, a system reader, an open-source tool, or an app like Pintail.
Does Claude Code have text to speech built in?
No. Its /voice command is dictation: you speak, it types your prompt. Anthropic’s help page says it plainly: in Claude Code, dictation is available but voice mode isn’t. The Claude apps can talk back; Claude Code can’t. What /voice does and doesn’t do. Speech out is the missing half. Here are the four ways I know to get it, with what each takes to set up, what it sounds like and where it breaks.
1. A Stop hook and the system voice (free)
Claude Code runs hooks at fixed moments. The Stop hook fires when a reply is done, and its input includes the reply’s text as last_assistant_message. Pipe that into say.
- Setup: about 20 lines of shell and some
jq. Ten minutes if you’ve written hooks before. - Sounds like: the macOS system voice. Clear, but robotic.
- Breaks when: the reply has markdown (it reads the stars and backticks), two tabs finish at once (they talk over each other), or Claude starts a background job (it says “started”, then talks again when the job is done).
I wrote up a working Stop hook and every way it breaks.
2. The Mac reads it for you (free, built in)
macOS has Spoken Content built in: select text, press a shortcut, it reads the selection.
- Setup: one checkbox in System Settings → Accessibility → Spoken Content.
- Sounds like: the system voice.
- Breaks when: the reply is long. It reads everything you select, and you have to go and select it.
3. A free plugin or open-source app
Aloud reads a short gist of each reply with a local voice and works with Codex too. flowcode is a native menu bar app that reads replies and takes dictation, all on your Mac. VoiceMode turns the session into a two-way voice conversation. As Claude Code plugins, claude-read-aloud adds a speaker button in VS Code, Heard’s claude-code-tts is the Stop hook already written, and claude-voice-notify says the repo name and a short line when a session is done.
- Setup: from a two-line
/plugin installto Homebrew, Python and a local model. - Sounds like: the system voice, or a local neural voice, usually Kokoro, which is much better.
- Breaks when: you want a voice or language the local model doesn’t have, or the tool’s maintainer moves on. They’re free, and several are good.
4. An app built for it: Pintail
Pintail is what I built for myself. It hooks into Claude Code, writes a short spoken version of each reply at the length you pick, and speaks it. The voice is a cloud voice on your own OpenRouter account or on Pintail credits, and English can use Kokoro on the Mac. A pill at the top of the screen shows which project is talking, with live captions.
- Setup: install the app, connect Claude Code, then paste an OpenRouter key or buy Pintail credits. English can use Kokoro on the Mac.
- Sounds like: a human-sounding cloud voice. First sound about half a second to a second and a half after the summary starts. English Kokoro runs on the Mac.
- Costs: $15 once, plus a cloud voice, usually ½–2 ¢ a reply.
- Breaks when: you’re offline. English Kokoro runs on the Mac, and on macOS 26 the summary can too. Other languages need the cloud voice.
What if I only want a sound when it’s done?
Then you don’t need speech. Claude Code already sends a desktop notification in iTerm2, Ghostty and Kitty, and a Notification hook can play a sound in any terminal. What a chime can’t tell you is which session finished, or what it did. Every way to get notified when Claude Code finishes.
Which one should you use?
One tab and free: the Stop hook or Spoken Content. Fully local and free: Aloud or flowcode. Talking, not just listening: VoiceMode. Several tabs, short summaries, a voice you can listen to all day: Pintail. The full comparison has a table.