Fish Audio MCP server for your AI

No. The connect form asks for two values: API key (required) and TTS model (optional) — there is no OAuth app, client secret, or redirect URL to register. Copy the key from the Fish Audio dashboard: fish.audio → Go to App → API Keys. The connection is locked to the one Fish Audio account that owns that key.

It can list and search with fishaudio_list_models, which covers both your own custom models and Fish Audio's public voice library, pull a single model with fishaudio_get_model, and change a model's metadata with fishaudio_update_model. It can also build a custom clone from your own audio samples with fishaudio_create_model, and remove a model by id with fishaudio_delete_model. That delete is permanent and runs as soon as the AI calls it, so tick Fish Audio on the consent screen only for clients that should manage models; read-only access, on the Free plan or for a teammate, rules it out.

Yes. fishaudio_voice_design builds a brand-new voice from a text prompt alone, with no audio upload, and it is fully supported through this connector. If you want a clone of a specific real voice instead, fishaudio_create_model is the path that takes your own audio samples. Cloning is not instant — a model finishes training before you can use it, and PopMCP can watch for "Voice clone ready" and "Voice clone failed" so the agent is not guessing at status.

fishaudio_text_to_speech_with_timestamps is the recommended way to synthesize speech through this connector, and it returns timestamps alongside the audio. fishaudio_text_to_speech is there for plain synthesis when you do not need them. The connect form also takes an optional TTS model value if you want to name one up front.

Yes. fishaudio_speech_to_text runs Fish Audio ASR and returns text with per-segment timestamps, so an agent can take a recording and hand back a segmented transcript. That pairs with synthesis in one instruction: transcribe a call, then read the summary back in one of your own voices.

PopMCP can watch for it. "Credit running low" fires when the pay-as-you-go balance drops below a level you choose, and "Credit ran out" fires when it hits zero, at which point new speech jobs will fail. "Plan allowance running low" tracks the remaining allowance on your subscription package separately, since that is not the same pool as pay-as-you-go credit. The agent can also read both on demand with fishaudio_get_api_credit and fishaudio_get_user_package.

No. PopMCP hosts it and gives you a Streamable HTTP MCP URL. There is nothing to deploy, no callback URL to register, and no infrastructure to keep running or patch.

Any client that supports a remote Streamable HTTP MCP server: Claude Desktop, Claude Web, Claude Code, ChatGPT custom connectors, Codex and the Codex CLI, OpenCode and Z.ai's ZCode, as well as editors such as Cursor, Windsurf and Zed. Your client signs in to PopMCP with OAuth; the provider credential never leaves the server.

Connect Fish Audio in PopMCP first, then run claude mcp add --transport http popmcp https://app.popmcp.com/mcp in a terminal. Inside Claude Code, run /mcp and sign in to PopMCP in the browser window it opens, ticking Fish Audio on the consent screen. The Fish Audio tools are then available in that session, and no Fish Audio credential is stored on your machine.

Yes. The Fish Audio catalog is small enough that all 11 tool definitions load into every conversation, so your AI never has to search for one before it can call it.

A tool is something your AI calls, and there are 11 here. A trigger is something PopMCP watches for you: 7 Fish Audio events that can start work on the PopMCP side. Triggers never enter your AI client's context, and none of them ask you to paste a callback URL into Fish Audio.