MCP agents that run the workspace you already have

Create a key in Settings → Agents. Then pick how the client talks to SuperKreator.

A Claude Skill is a file Claude installs, then it calls the API. Remote MCP is hosted: Claude, ChatGPT, or Cursor add the SuperKreator server and send the key. Local MCP is the same tools for a client that should not call the public URL. CLI and REST are the raw endpoints when you want a script.

The setup commands and the skill download are on the agents page. Whichever client you use, the operation list is the same.

A key can do the jobs exposed to agents: create or edit a product, save an AutoDM, schedule publishing, read orders, and the other catalog operations. It cannot invent a feature the dashboard does not have.

Keys are scoped. Rotate or delete a key that leaked. Hourly rate limits apply on the API.

The picture of Claude, Cursor, and your own agent is the idea. The live instructions are the four connect cards on /agents, not a separate agent marketplace inside the dashboard.

You can say "make a ₹499 reel pack and AutoDM anyone who comments LINK" in Claude instead of clicking through products and automations.

The result is still a draft in SuperKreator. You review it before it is public.

Cursor and other MCP clients can do the same jobs on a key you control, so a coding agent and a creator workspace are not two unrelated logins.

The dashboard is the source of truth. Use it when you are designing the page, checking a failed publish, or connecting Instagram for the first time.

Use an agent when the task is a sentence you would otherwise click twenty times: draft the product, draft the DM, queue the post, read today's orders.

There is no third-party bot store in the middle. The client you trust holds the key. If you want the full operation list, start at agents rather than a generic MCP directory.