supertonic3-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@supertonic3-mcp Say 'Hello, world!' in a cheerful voice"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
supertonic3-mcp
Local, on-device TTS for Claude & Cursor, powered by Supertonic 3. No API key. No cloud. An internal tool open-sourced by Halozen — we build AI compliance intelligence for construction.
Not affiliated with Supertone Inc.
Expose speak, list_voices, and list_expressions to Claude Desktop, Cursor, or any MCP client over STDIO.
Quick start (TTHW < 3 min)
git clone https://github.com/nextic-tech/supertonic3-mcp && cd supertonic3-mcp
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
# Optional: pre-download model for offline use (~400MB)
supertonic3-mcp preload
# Run MCP server (STDIO)
supertonic3-mcpCursor MCP config
Add to .cursor/mcp.json (or Cursor Settings → MCP):
{
"mcpServers": {
"supertonic3": {
"command": "/absolute/path/to/supertonic-tts/.venv/bin/supertonic3-mcp",
"args": []
}
}
}First server start downloads the Supertonic model into ~/.cache/supertonic3/ unless you ran preload first.
Related MCP server: Kokoro MCP Server
Tools
Tool | Description |
| Synthesize text to a WAV file; returns absolute path + metadata |
| Built-in voices ( |
| Inline tags ( |
speak parameters
text— 1–5000 characters; expression tags allowedvoice_id— optional (M1,F1, …)language— ISO 639-1 (en,ko,ja, …). For non-English text, always setlanguage=. Defaults toen.speed—0.7to2.0(SDK range)play— iftrue, plays audio on this machine viaafplay(macOS) oraplay(Linux). Unsupported on Windows.
WAV files are written to /tmp/supertonic_*.wav (macOS/Linux). Windows is not supported for synthesis output paths in v1.0.
Example return:
Audio saved to /tmp/supertonic_abc123.wav (1.4s, voice: M1, lang: en)Performance (this repo)
Measured on Apple M3, supertonic 1.3.1 — see benchmark/results.md.
Scenario | Median FSL |
Warm (model loaded) | ~0.82s |
Cold (new | ~0.81s |
FSL = time from synthesize() through WAV written (no streaming, no play=True).
Re-run: python benchmark/run.py
Offline use
supertonic3-mcp preloadDownloads ONNX weights atomically to ~/.cache/supertonic3/ and prints SHA256 checksums. After preload, synthesis works without network access.
Development
pip install -e ".[dev]"
pytestTests mock the Supertonic SDK (no network in CI).
Coming in v1.1
listen()— Whisper speech-to-text (pip install supertonic3-mcp[stt])SSE transport + Docker image for remote agents
PyPI publish workflow
License
MIT (this package). Supertonic SDK is MIT; model weights use OpenRAIL-M.
Disclaimer
AI-generated speech is not a substitute for certified safety, legal, or medical guidance. For demonstration purposes only.
Available Tools
3 toolslist_expressionsA
List inline expression tags (e.g. ) usable inside speak() text.
Inputs: no named parameters. Returns: result
list_voicesA
List built-in Supertonic voice styles.
Inputs: no named parameters. Returns: result
speakA
Convert text to speech and save a WAV file. Returns the absolute path and metadata.
Inputs: text (required), play, speed, language, +1 more. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Each tool has a clear, distinct purpose: listing voice styles, listing expression tags, and performing text-to-speech conversion. No overlap or ambiguity between the tools.
All tool names follow a consistent verb_noun pattern: list_voices, list_expressions, and speak. The naming is predictable and uniform.
With only three tools, the server is well-scoped for a text-to-speech domain. Each tool is essential and earns its place.
The tool set covers the core workflow: discovering available voices and expressions, then speaking text. No obvious gaps for the stated purpose.
Related MCP Connectors
Persistent memory for Claude Code and Cursor. Stop re-explaining your project every session.
Generate AI images, video, speech, music and presentations from Claude, ChatGPT and Cursor.
Text to speech for your AI. Your AI can send text to Doc Player to read it aloud. You will see a reader window with the text and you can control the playback sentence by sentence. Find an example here: https://documentplayer.com/connect-ai/
Synthesize short texts into WAV audio using offline formant voices in English, German and Spanish
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables voice interaction with Claude Code through local speech-to-text (Whisper) and text-to-speech (Supertonic), allowing verbal input/output without external API calls.2-
- AlicenseNot gradedqualityDmaintenanceProvides text-to-speech generation using the Kokoro-82M model, enabling AI assistants to generate voiceovers and audio content directly within Claude Desktop and Cursor.15Apache 2.0
- AlicenseNot gradedqualityDmaintenanceAdds text-to-speech capabilities to Cursor IDE, allowing your AI assistant to speak responses, summaries, and explanations out loud using OpenAI or ElevenLabs.1Apache 2.0

supertone-mcpofficial
AlicenseAqualityFmaintenanceMCP server for the Supertone TTS API. Generate natural speech, browse and preview the voice catalog, predict synthesis cost, and create cloned voices — directly from Claude Desktop, Cursor, or any MCP-compatible client. Supports Korean, English, Japanese, and 20+ other languages, with speed, pitch, and emotion-style control.1437 PyPI4MIT