mcp.film

mcp.film / Voice & Dialogue / Hume MCP Server

Hume MCP Server

Officialby Hume AIfreemiumverified Aug 17, 2026

Official Hume server for Octave TTS — emotionally expressive, acting-directed speech.

What it does

Hume's official MCP server drives Octave, a speech-language model that performs text rather than merely pronouncing it, with prompt-directed emotion and delivery. Ideal for film dialogue and narration where acting nuance matters; agents can iterate on takes, save designed voices to a library, and replay prior audio without regenerating.

Connect

Claude Code
claude mcp add hume -e HUME_API_KEY=YOUR_KEY -- npx @humeai/mcp-server
Claude Desktop (claude_desktop_config.json)
{
  "mcpServers": {
    "hume": {
      "command": "npx",
      "args": [
        "@humeai/mcp-server"
      ],
      "env": {
        "HUME_API_KEY": "YOUR_KEY_HERE"
      }
    }
  }
}
Cursor (.cursor/mcp.json)
{
  "mcpServers": {
    "hume": {
      "command": "npx",
      "args": [
        "@humeai/mcp-server"
      ],
      "env": {
        "HUME_API_KEY": "YOUR_KEY_HERE"
      }
    }
  }
}
Auth: api_key · env HUME_API_KEYhttps://platform.hume.ai

Tools you'll see

ttsplay_previous_audiolist_voicessave_voicedelete_voice

Field notes

Official but drifting toward dormant, and nothing moved between the last two checks: npm @humeai/mcp-server is still 0.3.0, published 8 Jan 2026, with no release since; repo last pushed 24 Jun 2026, ~9 stars, no LICENSE file in the tree even though npm metadata declares ISC. Five tools — a much narrower surface than ElevenLabs or Cartesia, and TTS-only, so pair it with something else for SFX and transcription. Set --workdir/-w (or WORKDIR) or audio lands in the system temp directory under /hume-tts; local playback wants ffplay or another CLI player on PATH. EMBEDDED_AUDIO_MODE (default false) returns audio inline instead, which is what you want in a client that cannot reach your filesystem.

Pairs well with