Agents

Rubikc is built to be worked on by agents. Almost everything in your Space — cubes, content, feeds, structure, imports, brand — can be changed by an AI assistant connected over MCP, not just through Admin.

You never have to use one. But the reach is deliberate: if you want an assistant to do the work, it can.

What an agent is good at

The honest answer is: reading a lot and writing first drafts.

• Auditing. “Which cubes are missing detail?” “Which products have only one image?” Work that is trivial to describe and tedious to do across a whole catalogue.
• Drafting. Producing first-pass content for cubes from a brief, which you then edit. Faster than a blank page, rarely right first time.
• Explaining. “Detail retention dropped last week — which cubes changed?” Agents are good at pulling several views together into an answer.
• Bulk tedium. The same small edit across many cubes.

What it is bad at

Taste, and anything where being wrong is expensive. An agent does not know what your brand sounds like unless you tell it, and it has no way of knowing that a particular product matters more than the others.

Tip

Your content playbook is where you write down the guidance an agent should follow. It is worth filling in properly before asking for content.

Working safely

An agent can change your Space. The good news is that the risky part is narrow: drafting content is safe because unpublished content is not visible, and publishing is the verb to be careful with.

Working safely with an agent covers what to delegate and what to keep for yourself.

Getting set up

What agents can doWhat is reachable, what is not, and working across systems.

Connect your agentTwo ways to connect Claude, Cursor or Codex.

Working safely with an agentWhat to let an agent change, and what to keep for yourself.