@alanse/fish-audio-mcp-server

Fish Audio MCP Server

An MCP (Model Context Protocol) server that provides seamless integration between Fish Audio's Text-to-Speech API and LLMs like Claude, enabling natural language-driven speech synthesis.

What is Fish Audio?

Fish Audio is a cutting-edge Text-to-Speech platform that offers:

• 🌊 State-of-the-art voice synthesis with natural-sounding output
• 🎯 Voice cloning capabilities to create custom voice models
• 🌍 Multilingual support including English, Japanese, Chinese, and more
• ⚡ Low-latency streaming for real-time applications
• 🎨 Fine-grained control over speech prosody and emotions

This MCP server brings Fish Audio's powerful capabilities directly to your LLM workflows.

Features

• 🎙️ High-Quality TTS: Leverage Fish Audio's state-of-the-art TTS models
• 🌊 Streaming Support: Real-time audio streaming for low-latency applications
• 🎨 Multiple Voices: Support for custom voice models via reference IDs
• 🎯 Smart Voice Selection: Select voices by ID, name, or tags
• 📚 Voice Library Management: Configure and manage multiple voice references
• 🔧 Flexible Configuration: Environment variable-based configuration
• 📦 Multiple Audio Formats: Support for MP3, WAV, PCM, and Opus
• 🚀 Easy Integration: Simple setup with any MCP-compatible client

Quick Start

Installation

You can run this MCP server directly using npx:

npx @alanse/fish-audio-mcp-server

Or install it globally:

npm install -g @alanse/fish-audio-mcp-server

Configuration

• Get your Fish Audio API key from Fish Audio

• Set up environment variables:

export FISH_API_KEY=your_fish_audio_api_key_here

• Add to your MCP settings configuration:

Single Voice Mode (Simple)

{
  "mcpServers": {
    "fish-audio": {
      "command": "npx",
      "args": ["-y", "@alanse/fish-audio-mcp-server"],
      "env": {
        "FISH_API_KEY": "your_fish_audio_api_key_here",
        "FISH_MODEL_ID": "speech-1.6",
        "FISH_REFERENCE_ID": "your_voice_reference_id_here",
        "FISH_OUTPUT_FORMAT": "mp3",
        "FISH_STREAMING": "false",
        "FISH_LATENCY": "balanced",
        "FISH_MP3_BITRATE": "128",
        "FISH_AUTO_PLAY": "false",
        "AUDIO_OUTPUT_DIR": "~/.fish-audio-mcp/audio_output"
      }
    }
  }
}

Multiple Voice Mode (Advanced)

{
  "mcpServers": {
    "fish-audio": {
      "command": "npx",
      "args": ["-y", "@alanse/fish-audio-mcp-server"],
      "env": {
        "FISH_API_KEY": "your_fish_audio_api_key_here",
        "FISH_MODEL_ID": "speech-1.6",
        "FISH_REFERENCES": "[{'reference_id':'id1','name':'Alice','tags':['female','english']},{'reference_id':'id2','name':'Bob','tags':['male','japanese']},{'reference_id':'id3','name':'Carol','tags':['female','japanese','anime']}]",
        "FISH_DEFAULT_REFERENCE": "id1",
        "FISH_OUTPUT_FORMAT": "mp3",
        "FISH_STREAMING": "false",
        "FISH_LATENCY": "balanced",
        "FISH_MP3_BITRATE": "128",
        "FISH_AUTO_PLAY": "false",
        "AUDIO_OUTPUT_DIR": "~/.fish-audio-mcp/audio_output"
      }
    }
  }
}

Environment Variables

Variable	Description	Default	Required
FISH_API_KEY	Your Fish Audio API key	-	Yes
FISH_MODEL_ID	TTS model to use (s1, speech-1.5, speech-1.6)	s1	Optional
FISH_REFERENCE_ID	Default voice reference ID (single reference mode)	-	Optional
FISH_REFERENCES	Multiple voice references (see below)	-	Optional
FISH_DEFAULT_REFERENCE	Default reference ID when using multiple references	-	Optional
FISH_OUTPUT_FORMAT	Default audio format (mp3, wav, pcm, opus)	mp3	Optional
FISH_STREAMING	Enable streaming mode (HTTP/WebSocket)	false	Optional
FISH_LATENCY	Latency mode (normal, balanced)	balanced	Optional
FISH_MP3_BITRATE	MP3 bitrate (64, 128, 192)	128	Optional
FISH_AUTO_PLAY	Auto-play audio and enable real-time playback	false	Optional
AUDIO_OUTPUT_DIR	Directory for audio file output	~/.fish-audio-mcp/audio_output	Optional

Configuring Multiple Voice References

You can configure multiple voice references in two ways:

JSON Array Format (Recommended)

Use the FISH_REFERENCES environment variable with a JSON array:

FISH_REFERENCES='[
  {"reference_id":"id1","name":"Alice","tags":["female","english"]},
  {"reference_id":"id2","name":"Bob","tags":["male","japanese"]},
  {"reference_id":"id3","name":"Carol","tags":["female","japanese","anime"]}
]'
FISH_DEFAULT_REFERENCE="id1"

Individual Format (Backward Compatibility)

Use numbered environment variables:

FISH_REFERENCE_1_ID=id1
FISH_REFERENCE_1_NAME=Alice
FISH_REFERENCE_1_TAGS=female,english

FISH_REFERENCE_2_ID=id2
FISH_REFERENCE_2_NAME=Bob
FISH_REFERENCE_2_TAGS=male,japanese

Usage

Once configured, the Fish Audio MCP server provides two tools to LLMs.

Tool 1: fish_audio_tts

Generates speech from text using Fish Audio's TTS API.

Parameters

• text (required): Text to convert to speech (max 10,000 characters)
• reference_id (optional): Voice model reference ID
• reference_name (optional): Select voice by name
• reference_tag (optional): Select voice by tag
• streaming (optional): Enable streaming mode
• format (optional): Output format (mp3, wav, pcm, opus)
• mp3_bitrate (optional): MP3 bitrate (64, 128, 192)
• normalize (optional): Enable text normalization (default: true)
• latency (optional): Latency mode (normal, balanced)
• output_path (optional): Custom output file path
• auto_play (optional): Automatically play the generated audio
• websocket_streaming (optional): Use WebSocket streaming instead of HTTP
• realtime_play (optional): Play audio in real-time during WebSocket streaming

Voice Selection Priority: reference_id > reference_name > reference_tag > default

Tool 2: fish_audio_list_references

Lists all configured voice references.

Parameters

No parameters required.

Returns

• List of configured voice references with their IDs, names, and tags
• Default reference ID

Examples

Basic Text-to-Speech

User: "Generate speech saying 'Hello, world! Welcome to Fish Audio TTS.'"

Claude: I'll generate speech for that text using Fish Audio TTS.

[Uses fish_audio_tts tool with text parameter]

Result: Audio file saved to ./audio_output/tts_2025-01-03T10-30-00.mp3

Using Custom Voice by ID

User: "Generate speech with voice model xyz123 saying 'This is a custom voice test'"

Claude: I'll generate speech using the specified voice model.

[Uses fish_audio_tts tool with text and reference_id parameters]

Result: Audio generated with custom voice model xyz123

Using Voice by Name

User: "Use Alice's voice to say 'Hello from Alice'"

Claude: I'll generate speech using Alice's voice.

[Uses fish_audio_tts tool with reference_name: "Alice"]

Result: Audio generated with Alice's voice

Using Voice by Tag

User: "Generate Japanese speech saying 'こんにちは' with an anime voice"

Claude: I'll generate Japanese speech with an anime-style voice.

[Uses fish_audio_tts tool with reference_tag: "anime"]

Result: Audio generated with anime voice style

List Available Voices

User: "What voices are available?"

Claude: I'll list all configured voice references.

[Uses fish_audio_list_references tool]

Result:
- Alice (id: id1) - Tags: female, english [Default]
- Bob (id: id2) - Tags: male, japanese
- Carol (id: id3) - Tags: female, japanese, anime

HTTP Streaming Mode

User: "Generate a long speech in streaming mode about the benefits of AI"

Claude: I'll generate the speech in streaming mode for faster response.

[Uses fish_audio_tts tool with streaming: true]

Result: Streaming audio saved to ./audio_output/tts_2025-01-03T10-35-00.mp3

WebSocket Real-time Streaming

User: "Stream and play in real-time: 'Welcome to the future of AI'"

Claude: I'll stream the speech via WebSocket and play it in real-time.

[Uses fish_audio_tts tool with websocket_streaming: true, realtime_play: true]

Result: Audio streamed and played in real-time via WebSocket

Development

Local Development

• Clone the repository:

git clone https://github.com/da-okazaki/mcp-fish-audio-server.git
cd mcp-fish-audio-server

• Install dependencies:

npm install

• Create .env file:

cp .env.example .env
# Edit .env with your API key

• Build the project:

npm run build

• Run in development mode:

npm run dev

Testing

Run the test suite:

npm test

Project Structure

mcp-fish-audio-server/
├── src/
│   ├── index.ts          # MCP server entry point
│   ├── tools/
│   │   └── tts.ts        # TTS tool implementation
│   ├── services/
│   │   └── fishAudio.ts  # Fish Audio API client
│   ├── types/
│   │   └── index.ts      # TypeScript definitions
│   └── utils/
│       └── config.ts     # Configuration management
├── tests/                # Test files
├── audio_output/         # Default audio output directory
├── package.json
├── tsconfig.json
└── README.md

API Documentation

Fish Audio Service

The service provides two main methods:

• generateSpeech: Standard TTS generation

  • Returns audio buffer
  • Suitable for short texts
  • Lower memory usage

• generateSpeechStream: Streaming TTS generation

  • Returns audio stream
  • Suitable for long texts
  • Real-time processing

Error Handling

The server handles various error scenarios:

• INVALID_API_KEY: Invalid or missing API key
• NETWORK_ERROR: Connection issues with Fish Audio API
• INVALID_PARAMS: Invalid request parameters
• QUOTA_EXCEEDED: API rate limit exceeded
• SERVER_ERROR: Fish Audio server errors

Troubleshooting

Common Issues

• "FISH_API_KEY environment variable is required"

  • Ensure you've set the FISH_API_KEY environment variable
  • Check that the API key is valid

• "Network error: Unable to reach Fish Audio API"

  • Check your internet connection
  • Verify Fish Audio API is accessible
  • Check for proxy/firewall issues

• "Text length exceeds maximum limit"

  • Split long texts into smaller chunks
  • Maximum supported length is 10,000 characters

• Audio files not appearing

  • Check the AUDIO_OUTPUT_DIR path exists
  • Ensure write permissions for the directory

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

• Fork the repository
• Create your feature branch (git checkout -b feature/AmazingFeature)
• Commit your changes (git commit -m 'Add some AmazingFeature')
• Push to the branch (git push origin feature/AmazingFeature)
• Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Fish Audio for providing the excellent TTS API
• Anthropic for creating the Model Context Protocol
• The MCP community for inspiration and examples

Support

For issues, questions, or contributions, please visit the GitHub repository.

Changelog

v0.6.0 (2025-01-03)

• Added multiple voice reference management system
• New tool: fish_audio_list_references to list configured voices
• Voice selection by name or tag in addition to ID
• Support for configuring multiple voices with metadata
• Added FISH_REFERENCES and FISH_DEFAULT_REFERENCE environment variables
• Enhanced voice selection with priority: ID > Name > Tag > Default

v0.5.4 (2025-01-03)

• Fixed zod version compatibility issues
• Resolved dependency conflicts between MCP SDK and Fish Audio SDK
• Verified local dev and build functionality

v0.5.3 (2025-01-03)

• Fixed missing zod dependency causing module resolution errors
• Improved compatibility when running via npx

v0.5.2 (2025-01-03)

• Fixed audio playback issue with FISH_STREAMING=true
• Fixed tilde (~) expansion in AUDIO_OUTPUT_DIR
• Improved stability by separating HTTP and WebSocket streaming

v0.5.1 (2025-01-03)

• Improved documentation formatting and clarity
• Updated environment variables table for better readability
• Made documentation more generic for all MCP clients

v0.5.0 (2025-01-03)

• Simplified environment variables: removed FISH_WEBSOCKET_STREAMING and FISH_REALTIME_PLAY
• WebSocket streaming now controlled by FISH_STREAMING
• Real-time playback now controlled by FISH_AUTO_PLAY
• Cleaner configuration with unified controls

v0.4.1 (2025-01-03)

• Added intelligent environment variable mapping
• FISH_WEBSOCKET_STREAMING defaults to FISH_STREAMING
• FISH_REALTIME_PLAY defaults to FISH_AUTO_PLAY
• Simplified configuration with smart defaults

v0.4.0 (2025-01-03)

• Refactored to use official Fish Audio SDK
• Improved WebSocket streaming implementation
• Fixed auto-play functionality
• Better error handling and connection stability
• Latency parameter now properly supported (normal/balanced)
• Cleaner codebase with SDK integration

v0.3.0 (2025-01-03)

• Added WebSocket streaming support for real-time TTS
• Added real-time audio playback during WebSocket streaming
• New parameters: websocket_streaming and realtime_play
• Support for both HTTP and WebSocket streaming modes
• Real-time player for immediate audio output

v0.2.0 (2025-01-03)

• Added automatic audio playback feature with auto_play parameter
• Added FISH_AUTO_PLAY environment variable for default behavior
• Support for cross-platform audio playback (macOS, Windows, Linux)
• HTTP streaming mode implementation

v0.1.2 (2025-01-03)

• Changed npm package name to @alanse/fish-audio-mcp-server

v0.1.1 (2025-01-03)

• Fixed directory creation error when running via npx
• Changed default audio output to user's home directory

v0.1.0 (2025-01-03)

• Initial release
• Basic TTS functionality
• Streaming support
• Environment variable configuration
• Multiple audio format support

Changelog

[0.6.1] - 2025-01-04

Fixed

• Fixed MCP JSON communication issues by replacing console output with logger utility
• Logger only outputs when DEBUG environment variable is set
• Prevents stdout/stderr messages from interfering with MCP JSON-RPC protocol