tts-mcp

MCP server that provides Text-to-Speech capabilities. It accepts text output from LLMs, synthesizes it using remote or local audio providers, and plays it directly through the host system's native speakers.

Features

• Direct Playback: Pipes synthesized audio to the host system natively.
• Provider Aggregation: Automatically exposes individual command tools for any configured TTS provider (fishaudio_tts, elevenlabs_tts, etc.).
• Persona Routing: Map specific voices and providers to semantic names to simplify LLM tool calls (speak_as_persona).
• Caching: Saves generated audio artifacts locally to an XDG cache directory.

Supported Providers

• FishAudio
• ElevenLabs
• Neets AI
• PlayHT
• Cartesia
• OpenAI
• Azure
• Local APIs

Installation

Requires Go 1.22+ to build from source via just init

Download the pre-compiled binaries matching your OS from the Releases page.

Configuration

tts-mcp requires an .env file in your standard OS configuration directory:

• Windows: %APPDATA%\tts-mcp\.env
• Linux: ~/.config/tts-mcp/.env
• macOS: ~/Library/Application Support/tts-mcp/.env

Run the included tts-mcp-config CLI to safely initialize this file, or populate it manually:

FISHAUDIO_API_KEY="<YOUR_API_KEY_HERE>"
ELEVENLABS_API_KEY="<YOUR_API_KEY_HERE>"

Setup

You can attach the server as a standard command-line MCP tool in your preferred AI editor.

Antigravity and Cursor

Add the executable to your MCP Servers configuration list:

• Name: tts-mcp
• Type: command
• Command: /absolute/path/to/extracted/tts-mcp

Claude Code and Claude Desktop

Append to your claude_desktop_config.json:

{
  "mcpServers": {
    "tts-mcp": {
      "command": "/absolute/path/to/extracted/tts-mcp"
    }
  }
}

Architecture

%%{init: {'theme': 'dark'}}%%
sequenceDiagram
    participant Client as IDE (Client)
    participant MCP as tts-mcp
    participant Config as XDG Config
    participant Provider as TTS API Runtime
    participant Audio as Native Audio Driver
    participant Cache as XDG Cache

    Client->>MCP: Call `speak_as_persona` (text, persona)
    MCP->>Config: Map persona to provider/voice_id
    MCP->>Provider: Request speech synthesis
    Provider-->>MCP: MP3/WAV Audio Stream

    par Playback
        MCP->>Audio: Buffer and pipe to host speakers
        MCP->>Cache: Save stream to persistent storage
    end

License

MIT