Readback

Voice output for Claude Code. Flip it on and Claude reads its replies aloud, so you can rest your eyes or keep half an ear on a long run instead of watching the terminal scroll. Flip it off and it stops immediately, mid-sentence if the phone rings.

• 🪟 Windows-first. Plays through built-in PowerShell audio, with zero external dependencies. (Most Claude voice tools are macOS-only.)
• 🎚️ Two providers. Inworld (hundreds of voices, cheap) and ElevenLabs, switchable in a click.
• 🖥️ GUI-tunable. A little control panel for voice, model, speed and expression, with a live voice picker and in-app key entry. No .env fiddling.
• ⚡ Streaming. Splits replies into sentences and starts talking on the first one, so audio kicks in fast even on long messages.
• 🔀 Multi-session. Running several Claude projects at once? Their replies queue and read in order instead of cutting each other off. Voice off still silences everything instantly.

Unofficial community tool. Not affiliated with, or endorsed by, Anthropic, Inworld, or ElevenLabs.

Already using it? See the changelog. 0.5.0 fixes the stop path at its root: stale player bookkeeping, concurrent saves, the process sweep, and long replies. It ships with regression tests that fail without those fixes. If you are on any 0.4.x, git pull to update.

---

How it works

Three cooperating pieces sharing one state file:

Piece	Role
MCP server	in-chat toggle: voice_on / voice_off / set_provider / set_voice / say / list_voices
Stop hook	the actual voice: auto-speaks each reply while enabled
Control panel	localhost:7717 web cockpit for provider / voice / model / tuning + live preview

The hook and MCP toggle work whether or not the panel is open.

Setup (Windows, Node 18+)

git clone https://github.com/Deltawerks/readback
cd readback
npm install
npm run panel        # opens the control panel in your browser

In the panel: pick a provider, paste that provider's API key, choose a voice, and hit ▶ to hear it. Then wire it into Claude Code:

npm run register     # writes .mcp.json + hooks-snippet.json for this folder

1. MCP server (the in-chat toggle). Auto-loads whenever you work in this folder. To get it in every project, run the claude mcp add line that register prints. That line uses user scope (-s user), which registers the server once for your whole account. It matters: claude mcp add defaults to local scope, which is only the project folder you ran it in, so without -s user you get the toggle tools in one folder while the hook goes on speaking in every other project.

2. Auto-speak hook (the part that actually talks). Open your Claude Code settings at C:\Users\<you>\.claude\settings.json. What you do next depends on what is already in that file.

No settings file yet: paste in the whole hooks-snippet.json that register just wrote, as the entire file.

A settings file with no hooks key: add the snippet's "hooks" block as one more top-level key, next to your existing settings.

A settings file that already has a hooks key: keep that key and put the snippet's "Stop" entry inside it, alongside the hooks you already have:

{
  "yourExistingSettings": "stay exactly as they are",
  "hooks": {
    "PreToolUse": [
      { "matcher": "Bash", "hooks": [ { "type": "command", "command": "node", "args": ["C:\\your\\existing\\hook.js"] } ] }
    ],
    "Stop": [
      { "hooks": [ { "type": "command", "command": "node", "args": ["C:\\path\\to\\readback\\hook\\stop-hook.js"], "timeout": 15 } ] }
    ]
  }
}

Never paste a second "hooks" key. An object with the same key twice is still valid JSON, and the parser keeps the last one, so every hook you had before is dropped, silently, with no error anywhere.

If you already have a "Stop" array of your own, the same rule applies one level down: add Readback's { "hooks": [ ... ] } entry to that existing array instead of adding a second "Stop" key.

Use the path from your generated snippet, not the one above. Then restart Claude Code, say "voice on", and the next reply should speak.

Optional: npm run shortcut drops a "Readback" icon on your Desktop that opens the panel in one click.

Quick smoke test, no Claude Code required:

npm run say "readback is online"
npm run voices        # list the active provider's voices

Day to day (and after a reboot)

Nothing to restart. Claude Code launches the MCP server itself, and the hook is just a line in its settings file, so both come back on their own after a reboot.

The control panel is optional and on-demand: a settings GUI, not a background service. Voice keeps working whether or not it's open, because the hook reads your saved settings from disk. Launch it (Desktop icon or npm run panel) when you want to switch voice, provider or speed, then close it again.

Using it

• In chat: say "voice on" and replies start speaking. "voice off" silences instantly, including whatever's playing right then.
• Multiple projects at once: their replies line up and read one after another instead of stomping each other. "voice off" (or the panel's Stop button) clears the whole queue at once, for when the phone rings.
• In the panel: switch provider, pick a voice, drag speed / expression, hit ▶ to preview. The panel's Speak/preview takes over immediately (it's you, at the keyboard); only the automatic per-reply speech queues.
• Voice off means off. The MCP say tool now refuses while voice is off, unless it is called with force: true, so Claude can't decide to talk over a mute you just pressed. The panel's Speak and preview buttons are unaffected: that is you at the keyboard, asking for it.

Keys are stored per-user outside the repo: %APPDATA%\Readback\secret.json on Windows (~/.config/readback/ elsewhere), so cloning into a shared or cloud-synced folder can't sync your key with it. Override the location with READBACK_STATE_DIR.

Working files (the log, the queue, the list of running players, the WAV chunks) live separately in %LOCALAPPDATA%\Readback (~/.cache/readback elsewhere), so they can't bloat a synced profile. Override that with READBACK_CACHE_DIR.

If you override either path, set it for every Readback process, not just one. The panel, the MCP server and each hook worker resolve these independently from the environment they were launched in. When they disagree, the toggle appears to work and controls nothing, because the half that speaks is reading different files than the half you clicked. npm run where prints the resolved paths; run it from the panel's environment and from a Claude Code session and compare. The panel also reports both at /health.

Replies are cleaned before speaking (code blocks dropped, links flattened, markdown/emoji stripped). Extremely long replies are capped with a spoken "the rest is on screen", but the cap is deliberately high so it acts as a backstop rather than clipping normal replies. Set READBACK_MAX_CHARS if you want a shorter ceiling on how long a read can run.

Upgrading from an earlier version? Your key and settings are copied to the new location automatically on first run. The originals are left in the repo's .readback/ folder (gitignored), so nothing is lost if you roll back. Delete that folder once you've confirmed things still work.

Notes & limits

• Windows only (PowerShell SoundPlayer playback). No STT / voice input.
• Speech is provider-agnostic WAV under the hood (Inworld LINEAR16; ElevenLabs PCM wrapped in a WAV header), so the streaming player never cares which provider you're on.
• The panel is loopback-only (127.0.0.1), rejects cross-origin requests, and bundles its own logo + fonts, so the page itself loads nothing from the internet.
• Synthesis does call out, by design: your cleaned, truncated reply text and your API key go to whichever provider you picked (Inworld or ElevenLabs) over HTTPS, and nowhere else. Readback has no servers, no telemetry, no analytics, and no update check.
• It costs money per character spoken. Readback defaults to the cheapest sensible model, Inworld tts-1.5-mini, at $0.005 per 1,000 characters ($5 per million); max is double that for a bit more richness. ElevenLabs costs 10 to 20 times as much: $0.05 per 1,000 characters on Flash and Turbo, $0.10 on eleven_v3 and eleven_multilingual_v2. At the default 12,000 character cap, one long reply is about 6 cents on Inworld mini and $0.60 to $1.20 on ElevenLabs. If you run ElevenLabs, set READBACK_MAX_CHARS to something shorter so a wall of text can't run up a dollar on its own. The panel shows the rate for the model you have selected. Worth knowing either way: if you add the hook to your global settings.json, it speaks in every project where voice is on, which adds up faster than you'd guess. Keep it per-project, or toggle voice off when you're not listening.
• Trouble? Check readback.log in %LOCALAPPDATA%\Readback on Windows (throwaway data is kept out of the roaming profile), or alongside the state dir otherwise.

License

MIT. See LICENSE.