vss-sound-mcp

MCP server for VFX Sound Studio (https://vss.ai.kr) — lets Claude Code, Claude Desktop, Cursor and any other MCP client generate layered game sound effects and monster voices from text and drop the WAV straight into your project.

Each sound is four recorded-foley layers (impact · body · texture · tail, or voice · texture · body · tail for creatures) generated by AI, judged by a listening model, then mixed offline with the same rules the web studio uses.

Install

1. Get an API key: sign in at https://vss.ai.kr → account menu (top right) → API 키 → 발급. New accounts get 50 free tokens, then 10 per day; packs at https://vss.ai.kr/pricing.
2. Add the server.

Claude Code

claude mcp add vss-sound -e VSS_API_KEY=vss_your_key -- npx -y vss-sound-mcp

Claude Desktop / Cursor / Windsurf / Cline / Roo Code / Gemini CLI — add to the tool's MCP settings JSON (claude_desktop_config.json, ~/.cursor/mcp.json or a project .cursor/mcp.json, ~/.codeium/windsurf/mcp_config.json, ~/.gemini/settings.json, or the MCP panel's config file):

{
  "mcpServers": {
    "vss-sound": {
      "command": "npx",
      "args": ["-y", "vss-sound-mcp"],
      "env": { "VSS_API_KEY": "vss_your_key" }
    }
  }
}

OpenAI Codex CLI (~/.codex/config.toml)

[mcp_servers.vss-sound]
command = "npx"
args = ["-y", "vss-sound-mcp"]
env = { VSS_API_KEY = "vss_your_key" }

VS Code / GitHub Copilot agent mode (.vscode/mcp.json)

{
  "servers": {
    "vss-sound": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "vss-sound-mcp"],
      "env": { "VSS_API_KEY": "vss_your_key" }
    }
  }
}

ChatGPT (app / web) cannot attach a local MCP server yet (it only takes remote connectors). Use Codex CLI with the same account, or the web studio.

Anything else that speaks MCP over stdio works the same way: command npx, args -y vss-sound-mcp, env VSS_API_KEY. Requires Node.js 18+ (npx ships with it). No npm account is needed.

Tools

| Tool | What it does | Tokens | |---|---|---| | generate_vfx_sound | Text → game VFX sound effect (27 types: impact, slash, burst, fire, ice, lightning, …). Writes one mixed WAV (optionally the 4 layers). ai_design:true lets an AI design the layers from your description; hits (1–8) makes combos, volleys and rapid fire. | 5 (8 with ai_design) | | generate_monster_voice | Creature + size + action → monster voice (roar, growl, attack, hit, death, idle, alert, move…). | 5 (8 with ai_design) | | list_vfx_types | The 27 effect types and the keywords that map a description to each. | 0 | | list_monster_options | Valid creatures / sizes / actions / moods / textures. | 0 | | get_balance | Remaining tokens and plan. | 0 |

Example prompts once the server is connected:

• "Make a lightning strike sound for my 2D game and save it to assets/sfx/lightning.wav."
• "Generate a large dragon roar, 3d, into sfx/monsters/."
• "번개가 땅에 내리꽂히는 효과음을 assets/sfx 에 넣어줘. 묵직하게."

Get better results

The MCP server cannot look at your sprite sheet or let you hand-tune layers the way the web studio does, so the description is everything. Give one specific line per effect: what happens on screen, the material, the motion, the weight.

• Good: "heavy boulder dropping onto packed earth, short and dull", "three quick consecutive sniper shots, sharp crack with a short whistle", "thick green poison fog spreading, bubbling hiss, no impact".
• Weak: "skill 3", "attack sound", "cool explosion".
• Also say the length in seconds, the number of hits (combos / volleys), and 2D or 3D. If a result misses, refine the same description and call again.

Notes

• Output: 44.1 kHz 16-bit mono WAV. dimension:"2d" trims sub-bass for pixel/2D games; "3d" keeps full low end.
• Descriptions can be Korean or English. Name a real material and action ("boulder dropped on packed earth, short") — that is what the sound model renders best.
• The server never stores your key; it is read from VSS_API_KEY (or --key) and sent only to vss.ai.kr's functions.
• Rate limit: 8 requests per minute per account.

MIT © TouchErica