Skip to content

What MCP is

MCP — the Model Context Protocol — is an open standard for connecting AI assistants to systems they can read and act on. An assistant that speaks MCP can work with any service offering an MCP server, without anyone building a custom integration for that pairing.

Rubikc offers an MCP server. That is why the same setup works in Claude, Cursor and Codex.

An MCP URL points at a server scoped to your account. When an assistant connects, it logs in as you and can see what you can see.

The server exposes a set of tools — named operations the assistant can call. It cannot do anything to your Space that is not one of these.

A connection is not anonymous. Each one has a name, and Admin lists them with how they connected — through an assistant’s own connector, or a URL you created — and when each was last used.

That means your assistants are distinguishable rather than interchangeable. You can see which are still in use, and revoke one without disturbing the others.

See Connect your agent for the two ways to set one up.

Area What it covers
Spaces List Spaces and read their overall shape
Cubes Create, update, search and publish cubes
Content Blocks, versions, ordering, shared content, overlays. Assistants can look up every block type and its schema before writing.
Structure Navigation and structures
Feed Feed configuration
Sources Imports and source material
Files Uploads and media
Brand Brand and appearance settings
Analytics Storefront and cube analytics, sessions, surveys
Onboarding Onboarding progress and context

The server also hands assistants your content playbook, so an agent works from your written guidance rather than inventing an approach.

An agent has your access, not more. If you cannot reach a Space, neither can it. If you lose access to a Space, the agent loses that access immediately.

You can revoke a connection at any time from Settings › MCP. Revoking takes effect immediately, and affects only that connection.