---
name: vibegen-mcp-install
description: Set up the VibeGen MCP in your AI. Paste this file into your AI chat (Claude Code, Claude Desktop, Cursor, anything that speaks MCP) and say "set this up" — your AI downloads VibeGen, installs it, and wires it up. No pip, no config, no tokens to paste.
---

# VibeGen MCP — drag this into your AI and say "set it up"

VibeGen lets your AI drive the VibeDNA generative art studio: read the live canvas, switch engines, move every slider, and (on Ultra) rewrite the generative math live. This file is all your AI needs — it does the whole install.

## AI: set this up for the user

1. Download and unzip to `~/vibegen-mcp/`:
   https://vibedna.ai/api/free-download?slug=vibegen
2. Install deps: `pip install -r ~/vibegen-mcp/requirements.txt`
3. Add this MCP server to the user's config (`.mcp.json` for Claude Code, or their MCP client), using the ABSOLUTE path to server.py:

```json
{
  "mcpServers": {
    "vibegen": {
      "type": "stdio",
      "command": "python",
      "args": ["<ABSOLUTE_PATH>/vibegen-mcp/server.py"],
      "env": {
        "STUDIO_URL": "http://localhost:3000",
        "STUDIO_TOKEN": ""
      }
    }
  }
}
```

### Where the studio runs — your machine, never ours

The studio renders in **your own browser**: the art and the video bake happen on your GPU, client-side, so nothing renders on our servers. Point `STUDIO_URL` at wherever you run the studio:

- **`http://localhost:3000`** (default) — run the studio on your own machine.
- **your own VPS** — run it on a box you control and use that URL.
- **the hosted studio** at `https://vibedna.ai/studio` — open it in your browser; the render still happens locally in that browser, on your GPU, not ours.

Your choice. The MCP just drives whichever one you point it at.

4. Tell the user to close and reopen the tab so the MCP loads.
5. Confirm it works: call `get_state` (or say "get the VibeGen studio state").

## Tiers

No token = the free tier: read the canvas and play with every engine and slider. Creating and rewriting engines needs Ultra.

Subscribers get their token baked in automatically: go to **vibedna.ai/studio**, click **Set up in your AI**, and drop the file it gives you into your AI chat instead of this one. Same steps, your key already filled in.

## What your AI can do once it's wired

| Say to your AI | Tool |
|---|---|
| "get the studio state" | `get_state` |
| "make it warmer / slower / denser" | `set_params` |
| "switch to the plasma engine" | `set_engine` |
| "save this as a preset" | `write_save` / `apply_save` |
| "show me the engine's math" | `read_engine_source` |
| "rewrite the shader so the flow spirals" *(Ultra)* | `write_engine_source` |
| "snapshot what it looks like now" | `snapshot_canvas` |

## Trouble

- **Cannot reach studio** — the studio host is wrong or unreachable; check `STUDIO_URL`.
- **401 unauthorized** — that action needs Ultra. Get your key at vibedna.ai/studio → Set up in your AI.
- **MCP doesn't appear** — close and reopen the tab; the MCP list is read at session start.

## Support

`dev@vibedna.ai` — a human reads everything.

— VibeDNA, 2026