Skip to main content
Glama
yavru421

speech-mcp-server

by yavru421
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

speech-mcp-server

Speech MCP Server Banner

Local Stdio Model Context Protocol (MCP) server wrapping the Kokoro ONNX Neural TTS Engine with real-time Voice Vector Algebra, Central Wisconsin Vernacular Lexicon Overrides, and thread-clamped zero-VRAM CPU execution.

"Neural voice vectors mixed mathematically. Zero VRAM overhead. Pure local performance."

🔊 Key Architectural Features

  • Voice Vector Algebra Engine: Mix multiple voice profiles using linear vector algebra directly in Python: $$\text{Voice} = (\text{am_adam} \times 0.65) + (\text{bm_lewis} \times 0.30) + (\text{am_michael} \times 0.05)$$

  • Central Wisconsin Vernacular Dict: Built-in phonetic lexicon mapper enforcing regional dialect pacing and word pronunciations (bag $\to$ bayg, roof $\to$ ruff, creek $\to$ crick, know'm'sayin).

  • Zero GPU VRAM Overhead: Operates entirely on CPU via ONNX Runtime Execution Provider. Keeps GPU completely idle (0% load, 0 MB VRAM).

  • Thread-Clamped CPU Execution: Throttled to 2 worker threads (OMP_NUM_THREADS = 2) to eliminate CPU spikes while maintaining sub-200ms latency.

  • Kokoro ONNX Engine: High-fidelity 24kHz neural TTS with multiple base voice profiles (am_adam, af_bella, bm_lewis, am_michael).

  • MCP Stdio Transport: Full Model Context Protocol compatibility for AI agents and LLM tool integration (/speak).

Related MCP server: STT2TTS MCP

🚀 Quickstart

Prerequisites

  • Node.js v18+

  • Python 3.10+ with kokoro-onnx, onnxruntime, soundfile, numpy

Installation & Build

git clone https://github.com/yavru421/speech-mcp-server.git
cd speech-mcp-server
npm install
npm run build

Usage

python synth.py "Lexicon overrides active for DuckDB and WASAPI." "am_adam*0.65 + bm_lewis*0.30" "0.94"

Configuration (MCP Client)

Add to your mcpServers configuration:

{
  "speech-mcp-server": {
    "command": "node",
    "args": ["<path_to_repo>/build/index.js"]
  }
}

🔒 License

MIT

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivitySlowing
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • STT2TTS MCP

    pygodzilla
    A
    license
    Not graded
    quality
    D
    maintenance
    Local-first speech-to-text and text-to-speech MCP server. Hot-swappable engines via config.yaml — no code changes, no API keys required.
    2
    MIT

  • MCP Voice Soundboard

    mcp-tool-shop-org
    A
    license
    Not graded
    quality
    D
    maintenance
    A text-to-speech MCP server with 48 voices across 9 languages, supporting emotion spans, SFX tags, and multi-speaker dialogue. Deployable via a single npx command with built-in guardrails and swappable backends.
    MIT

  • kokoro-tts-mcp

    scottschram
    A
    license
    Not graded
    quality
    C
    maintenance
    Text-to-speech MCP server using the Kokoro-82M model accelerated with MLX on Apple Silicon, enabling local Claude and Codex clients to speak text aloud and convert text to audio.
    4
    MIT