Fish MCP server

MCP Fish Server

<!-- mcp-name: io.github.AceDataCloud/mcp-fish -->

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

Documentation

<!-- canonical-documentation --> Documentation

License

MIT — see LICENSE at the repository root.