{
  "slug": "fish-audio",
  "name": "Fish Audio MCP",
  "vendor": "Fish Audio",
  "official": true,
  "category": "voice-speech",
  "featured": false,
  "tagline": "Hosted Fish Audio server: expressive TTS, cloning, transcription, plus quoted image, video and lip-sync jobs.",
  "description": "Fish Audio's official hosted MCP server connects an agent to your Fish Audio account over OAuth: search the voice library and clone voices, generate speech steered with inline tags like [whispering] or [excited], transcribe local or uploaded audio, and — since September and October 2026 — generate, edit and upscale images and run text-to-video, image-to-video, lip sync and talking avatars behind a free quote step. For a filmmaker-agent the draw is temp dialogue and scratch VO with emotional direction written into the script line, then a lip-synced or avatar pass on the same account, all billed against the plan you already have.",
  "capabilities": [
    "tts",
    "voice-library",
    "voice-cloning",
    "emotion-tags",
    "stt",
    "text-to-image",
    "image-editing",
    "upscaling",
    "text-to-video",
    "image-to-video",
    "lip-sync",
    "talking-head",
    "cost-estimation"
  ],
  "tools_sample": [
    "search_voices",
    "create_voice_clone",
    "text_to_speech",
    "speech_to_text",
    "list_media_models",
    "estimate_image_generation",
    "estimate_video_generation",
    "generate_image",
    "generate_image_edit",
    "generate_video",
    "get_generation_status",
    "get_generation_result",
    "studio_generate",
    "studio_export"
  ],
  "install": {
    "claude_code": "claude mcp add --transport http fish-audio https://api.fish.audio/mcp",
    "remote_url": "https://api.fish.audio/mcp",
    "stdio_command": null
  },
  "auth": {
    "type": "oauth",
    "env_var": null,
    "key_url": "Browser OAuth with your Fish Audio account; you pick the team the connection works in"
  },
  "pricing": "freemium",
  "links": {
    "repo": null,
    "docs": "https://docs.fish.audio/overview/mcp",
    "site": "https://fish.audio"
  },
  "added": "2026-09-26",
  "verified": "2026-10-26",
  "caveats": [
    "Hosted only and closed source, and the tool names come from the announcement and docs, not a live tools/list.",
    "Every image or video job is quote, approve, then submit, and the server reprices at submission; if the price moved, quote again.",
    "Video, lip sync and avatars need Plus or higher, and the free tier is personal, non-commercial use only.",
    "Generated outputs come back as permanent URLs, so assume anything you generate is reachable by link."
  ],
  "notes": "Hosted only and closed source — no npm or PyPI package, no public server repo. Re-checked 2026-10-26: endpoint, Claude Code line and OAuth unchanged, but the surface grew. Correction to the earlier note here: Fish Audio's 7 Oct 2026 announcement adds image upscaling, lip sync and talking avatars, and the docs now list voice cloning, Story Studio (studio_generate, studio_export), an asset library and account tools. Lip sync and avatars run through generate_video rather than tools of their own. The tool names above come from that post and the docs page, not a live tools/list. Every image or video job goes discover, free estimate, your approval, then submit; the server reprices at submission and rejects the job if the price moved, so the agent has to quote again. Failed generations are refunded. Billing draws from your plan's package credits, not developer API credits; speech is 1 credit per UTF-8 byte and is charged on run with no quote. Video, lip sync and avatars need Plus ($11/mo promo, $15 list) or higher, Story Studio export needs a paid plan, and Fish publishes no fixed media price list — the estimate is the price; the free tier gets a daily image allowance and is personal, non-commercial use only, so anything headed for a paying client needs a paid plan. Connections are scoped to one team. Local files sent for transcription are uploaded to a temporary slot deleted after 7 days; generated outputs come back as permanent URLs, so assume anything you generate is reachable by link. Do not confuse it with Fish Audio's separate read-only docs MCP at docs.fish.audio/mcp, or with the community wrappers (da-okazaki/mcp-fish-audio-server, AceDataCloud/FishMCP), which are not maintained by Fish Audio.",
  "evidence": "endpoint_answered",
  "last_probe": {
    "date": "2026-11-10",
    "result": "auth_required",
    "protocol_version": null,
    "tool_count": null,
    "anonymous_tools_list": false,
    "auth_required": true,
    "oauth_metadata": true,
    "consecutive_failures": 0,
    "last_ok": "2026-11-10"
  },
  "package_check": null
}