MCP Fish Server
A Model Context Protocol (MCP) server for Fish Audio TTS (Text-to-Speech) via the AceDataCloud platform. Generate natural-sounding speech and explore the Fish voice model library.
Features
- High-quality TTS: Generate speech from text via Fish Audio models
- Voice library: Browse, search, and fetch metadata for Fish voice models
- Asynchronous tasks: Submit generation tasks and poll for results
- Batch task lookup: Query multiple task results in one call
One-shot voice cloning
Pass one public HTTPS reference audio URL plus its exact transcript. This conditions only the current TTS request and does not create a reusable voice model:
fish_generate_audio(
text="New speech in the referenced voice",
reference_audio_url="https://cdn.acedata.cloud/reference.mp3",
reference_text="The exact words spoken in the reference audio",
)
Use reference_id for saved or public voices, and the one-shot reference fields for a temporary voice. Do not combine them. Reference audio supports MP3/WAV and should be 10–270 seconds. Billing remains based on the target text's UTF-8 byte count.
Installation
pip install mcp-fish
Configuration
Set your AceDataCloud API token:
export ACEDATACLOUD_API_TOKEN=your_token_here
Get your token from https://platform.acedata.cloud.
Usage
stdio mode (default)
mcp-fish
HTTP mode
mcp-fish --transport http --port 8000
Tool Reference
| Tool | Description |
|---|---|
fish_generate_audio | Generate speech from text via a Fish voice model |
fish_list_models | List available Fish voice models |
fish_get_model | Fetch metadata for a specific Fish voice model |
fish_get_task | Get the status / result of a generation task |
fish_get_tasks_batch | Batch-fetch the status / result of multiple tasks |
fish_get_usage_guide | Get the API usage guide |
Documentation
License
MIT — see LICENSE at the repository root.