better-tts-mcp 0.2.0

A feature-rich MCP server for Microsoft Edge Text-to-Speech, supporting voice listing, speech synthesis with subtitles, parameter tuning, and batch processing.

better-tts-mcp

English | 中文

A feature-rich MCP (Model Context Protocol) server for Microsoft Edge Text-to-Speech. No API key required. 300+ voices in 50+ languages. Zero configuration.

---

Features

• List Voices — Browse 300+ voices in 50+ languages, filter by language and gender
• Text-to-Speech — Convert text to MP3 with customizable rate, volume, and pitch
• Subtitles — Generate SRT subtitle files alongside audio
• Batch Processing — Synthesize multiple texts in one call
• Multi-Voice Synthesis — Mix multiple voices in one output MP3 using [voice]text markers

Quick Start

Installation

# Using uvx (recommended)
uvx better-tts-mcp

# Or install via pip
pip install better-tts-mcp

Claude Code

claude mcp add edge-tts -- uvx better-tts-mcp

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "edge-tts": {
      "command": "uvx",
      "args": ["better-tts-mcp"]
    }
  }
}

---

Available Tools

list_voices

List available Edge TTS voices with optional filtering.

Parameter	Type	Required	Description
language	string	No	Filter by language code, e.g. "zh", "en"
gender	string	No	Filter by gender: "Male" or "Female"

text_to_speech

Convert text to speech and save as MP3.

Parameter	Type	Required	Default	Description
text	string	Yes	—	Text to convert
voice	string	No	zh-CN-XiaoxiaoNeural	Voice name
rate	string	No	—	Speed, e.g. "+50%", "-30%"
volume	string	No	—	Volume, e.g. "+20%", "-50%"
pitch	string	No	—	Pitch, e.g. "+10Hz", "-5Hz"
output_dir	string	No	. (current directory)	Output directory

text_to_speech_with_subtitles

Same parameters as text_to_speech. Generates both MP3 and SRT files.

batch_text_to_speech

Convert multiple texts to speech, each saved as a separate MP3 file.

Parameter	Type	Required	Default	Description
texts	string[]	Yes	—	List of texts to convert
voice	string	No	zh-CN-XiaoxiaoNeural	Voice name (shared for all texts)
rate	string	No	—	Speed (shared)
volume	string	No	—	Volume (shared)
pitch	string	No	—	Pitch (shared)
output_dir	string	No	. (current directory)	Output directory

text_to_speech_multi_voice

Convert text with voice markers into one merged MP3 file.

Format example:

[zh-CN-XiaoxiaoNeural]你好。[en-US-AriaNeural]Hello.

Parameter	Type	Required	Default	Description
text	string	Yes	—	Input text with optional [voice] markers
default_voice	string	No	zh-CN-XiaoxiaoNeural	Voice used when no marker is present
rate	string	No	—	Speed adjustment
volume	string	No	—	Volume adjustment
pitch	string	No	—	Pitch adjustment
output_dir	string	No	. (current directory)	Output directory

---

MCP Integration Contract (for external callers)

This server follows the MCP tool-call flow used by standard MCP clients:

• initialize -> capability negotiation
• tools/list -> discover tool names and JSON schemas
• tools/call -> invoke a tool by name with arguments

Transport:

• Default transport is stdio (entrypoint: better-tts-mcp)

Tool call example (tools/call)

{
  "method": "tools/call",
  "params": {
    "name": "text_to_speech",
    "arguments": {
      "text": "Hello from MCP",
      "voice": "en-US-AriaNeural",
      "rate": "+0%",
      "volume": "+0%",
      "pitch": "+0Hz",
      "output_dir": "./outputs"
    }
  }
}

Typical successful result text:

Audio saved to: /absolute/path/to/outputs/20260307_123456_Hello_from_MCP.mp3

Return-value conventions

• All tools return human-readable text in content.
• All tools also return machine-readable structuredContent with stable fields (ok, message, etc.) for robust LLM/tooling integration.
• Paths returned by synthesis tools are absolute paths.
• text_to_speech_with_subtitles returns two lines (audio path + subtitle path).
• batch_text_to_speech returns a summary header plus one output path per item.

Error behavior

• Malformed requests (JSON-RPC or schema-level) are protocol errors.
• Input validation and business/runtime failures are returned as tool execution errors with isError: true, so LLM clients can self-correct and retry.
• Error payloads include error_code, error_message, and retryable in structuredContent.

Common error codes:

• EMPTY_TEXT
• EMPTY_TEXTS
• EMPTY_TEXT_ITEM
• EMPTY_TEXT_SEGMENTS
• INVALID_RATE
• INVALID_VOLUME
• INVALID_PITCH
• SYNTHESIS_FAILED

Stability notes for API consumers

• Tool names are treated as stable API surface:
  • list_voices
  • text_to_speech
  • text_to_speech_with_subtitles
  • batch_text_to_speech
  • text_to_speech_multi_voice
• New optional parameters may be added in future versions without breaking existing callers.

---

Requirements

• Python >= 3.10
• Internet connection (for accessing Microsoft Edge TTS service)

License

MIT