Discord TTS Attacher

An OpenClaw plugin that turns your assistant’s Discord text replies into a spoken MP3 and posts it as a normal file attachment in a follow-up message. It uses Microsoft’s Edge neural voices (via node-edge-tts).

This is not Discord’s built-in “text-to-speech” in the client, and it is separate from OpenClaw’s other audio delivery paths. Listeners get a regular message with an MP3 they can play in Discord or download.

Synthesis runs in a separate Node child process (worker.mjs) so Edge TTS work stays off the gateway’s main event loop.¹

What you need

• OpenClaw running with the Discord channel set up
• Node.js 18 or newer
• Outbound internet access (synthesis goes through Microsoft’s Edge TTS service)

Install

Pick one way to install the plugin, then complete the shared steps below.

1. Install the files

ClawHub — from ClawHub, using the OpenClaw CLI (package discord-tts-attacher):

openclaw plugins install clawhub:discord-tts-attacher

OpenClaw places the plugin under your extensions layout; you do not copy folders by hand.

Manual — copy the plugin tree so the folder name is exactly discord-tts-attacher:

~/.openclaw/extensions/discord-tts-attacher/

A release zip or ClawHub package includes plugin sources, package.json, openclaw.plugin.json, LICENSE, CHANGELOG.md, and README.md—it does not include a vendored copy of node-edge-tts inside the archive. Instead, package.json declares node-edge-tts under dependencies, so when you run npm install, npm downloads and installs it from the registry into node_modules (you need network access to npm for that step). openclaw is an optional peer dependency so you usually do not install it in the extension folder; the SDK is loaded from the running gateway or OPENCLAW_PACKAGE_ROOT.

cd ~/.openclaw/extensions/discord-tts-attacher
rm -rf node_modules
npm install

The plugin finds the openclaw package root from the running gateway (walking upward from the host process’s argv[1]), from OPENCLAW_PACKAGE_ROOT, or—if you install openclaw here—via openclaw/package.json. It loads definePluginEntry from dist/plugin-sdk/plugin-entry.js, and sendMessageDiscord from dist/extensions/discord/runtime-api.js on OpenClaw 2026.4+, or from dist/plugin-sdk/discord.js on older installs. You do not need npm install openclaw in the extension folder unless you want that fallback.

If you used ClawHub and dependencies look wrong, run the same rm -rf node_modules && npm install in that plugin directory.

2. Enable in openclaw.json

• Add discord-tts-attacher to plugins.allow.
• Add an entry for discord-tts-attacher under plugins.entries (see the example below).

3. Restart

Restart the OpenClaw gateway so it loads the plugin.

Example openclaw.json snippet

"plugins": {
  "allow": ["discord", "discord-tts-attacher"],
  "entries": {
    "discord": { "enabled": true, "config": {} },
    "discord-tts-attacher": {
      "enabled": true,
      "config": {
        "voice": "en-US-AndrewNeural",
        "outputDir": "~/.openclaw/workspace/TTS",
        "maxTextLength": 3500
      }
    }
  }
}

Adjust paths for your machine (absolute paths are fine).

Configuration

All options go under plugins.entries.discord-tts-attacher.config. Only the keys below are supported; extra keys may be rejected when OpenClaw loads your config.

Timeouts

How long synthesis may run and how long the gateway waits for the finished MP3 both scale with reply length (longer text gets more time). The options in the table control those limits.

Option	Type	Default	What it does
enabled	boolean	true	Turn the plugin on or off.
voice	string	en-US-AndrewNeural	Edge neural voice name (e.g. en-US-AriaNeural).
outputDir	string	see below	Where temporary MP3s, job markers, and worker.log are written. Relative paths are resolved by OpenClaw.
channelAllowlist	array of strings	(empty = all channels)	If set, only those guild channel IDs (snowflakes) get TTS. Direct messages and other targets are not limited by this list.
maxTextLength	integer	12000	Skip synthesis if the reply text is longer than this.
sendDedupeMs	integer	4000	Ignore a second synthesis for the same target and text within this many milliseconds; use 0 to disable.
debounceMs	integer	3500	How long to wait after the last chunk of a streamed reply before starting synthesis.
pickupIntervalMs	integer	2000	How often the gateway polls for worker completion (MP3 + done marker).
pickupTimeoutMs	integer	60000	Minimum pickup wait (ms). Actual wait is max of this and synthTimeoutMs + pickupIntervalMs + 2000 (see above).
synthTimeoutBaseMs	integer	4000	Base milliseconds in the synthesis timeout estimate.
synthTimeoutPerCharMs	number	10	Milliseconds per character added to the synthesis timeout estimate.
synthTimeoutJitterMs	integer	5000	Extra buffer (ms) added to the synthesis timeout estimate.
synthTimeoutMaxMs	integer	120000	Upper cap (ms) for the synthesis timeout passed to the worker and node-edge-tts.

Default outputDir: $OPENCLAW_STATE_DIR/workspace/TTS, or ~/.openclaw/workspace/TTS if that variable is unset.

Environment variables

Variable	Purpose
DISCORD_TTS_OUTPUT_DIR	Overrides the default output directory when outputDir is not set in config.
OPENCLAW_STATE_DIR	Used when building the default TTS directory.
OPENCLAW_PACKAGE_ROOT	Absolute path to the openclaw npm package root (directory whose package.json has "name": "openclaw" and which contains dist/plugin-sdk/plugin-entry.js, with Discord send in dist/extensions/discord/runtime-api.js or legacy dist/plugin-sdk/discord.js). Use this if the gateway does not resolve the package via argv[1] walk and you have not installed the optional openclaw package inside the extension.

Choosing a voice

Voices use Microsoft’s naming pattern, for example en-US-AndrewNeural or en-GB-SoniaNeural. Search for “Edge TTS voices list” or use tools that ship with edge-tts / node-edge-tts to list available voices.

Privacy and files on disk

The plugin writes short-lived synthesis files and logs under outputDir (MP3s, completion markers, worker.log). worker.log is rewritten on each update so it only keeps entries from roughly the last 24 hours (older lines are dropped). Do not share that folder publicly if logs could expose sensitive routing details. Do not post or share your full openclaw.json (it can contain tokens and account information).

Maintainers

If you have the complete source repository (not only a minimal plugin package), open the development guide next to the plugin sources for release builds, ClawHub publishing, where openclaw.plugin.json defines the config schema, and the exact timeout math used in code. Use npm run release and npm run check-version from the repository root.

License

MIT License

Footnotes

1. The plugin uses Node’s child_process.spawn with process.execPath and a fixed path to the bundled worker.mjs, passing only a JSON job payload (text, voice, paths, timeouts). No shell is invoked (no shell: true, no arbitrary commands). Registries or scanners that flag child_process (for example ClawHub) are seeing this intentional worker offload, not generic shell execution. ↩