プラットフォーム

MCP server and CLI

Work with the studio from Claude, Cursor, any MCP client, or the terminal. It runs the same models at the same prices as the studio, on one API key that acts only on your workspace.

1. Mint a key

In the studio, open Settings → API keys and create a key. It is shown once; copy it then. A key can generate and read as an editor but cannot change billing, seats or keys, and you can revoke it at any time. Open Settings

2. Connect an MCP client

The endpoint speaks MCP over streamable HTTP. Send JSON-RPC to it with the key as a bearer token:

POST https://studio.ai-sns.io/mcp
Authorization: Bearer sk_studio_xxxx
Content-Type: application/json

A client that takes a remote server URL, such as Claude Code or Cursor, is configured like this:

{
  "mcpServers": {
    "studio": {
      "type": "http",
      "url": "https://studio.ai-sns.io/mcp",
      "headers": {
        "Authorization": "Bearer sk_studio_xxxx"
      }
    }
  }
}

A client that only runs local stdio servers, such as Claude Desktop, uses the bridge file instead; it forwards every message to the endpoint above:

curl -fsSL https://studio.ai-sns.io/cli/studio-mcp.js -o studio-mcp.js

{
  "mcpServers": {
    "studio": {
      "command": "node",
      "args": [
        "/path/to/studio-mcp.js"
      ],
      "env": {
        "STUDIO_API_KEY": "sk_studio_xxxx"
      }
    }
  }
}

3. Or install the command line

One file, no dependencies, Node 22 or newer. It prints the cost and asks before spending anything.

curl -fsSL https://studio.ai-sns.io/cli/studio.js -o /usr/local/bin/studio && chmod +x /usr/local/bin/studio
studio login          # paste the key; it is stored in ~/.studio/config.json (mode 600) and never shown
studio models         # the catalogue with its real prices
studio generate --model seedance-2.5 --prompt "a barista pouring latte art" --duration 5 --out clip.mp4

The tools

Every call is priced, checked and charged by the same code as the studio, so a generation costs exactly what it costs in the browser, and a model pass that covers it makes it free here too. A failed or rejected generation is refunded once. Calls are limited to 120 a minute per key. APIリファレンス