Security

How access control works in KBrain

Access control in KBrain is built on authentication, per-user MCP keys, brain visibility (private, public, open, organization), and the MCP security model. Here is who can query what, and why.

Connect your knowledge over MCP

Connect via MCP

Knowledge is only safe to connect to AI if you control who can reach it. Access control in KBrain rests on a few clear mechanisms: authenticated accounts, a per-user MCP key, and a visibility setting on every brain. Together they decide who can query what. This article walks through the model.

The access path

Access path - 01

From request to answer

Every query passes through authentication and a visibility check before any knowledge is returned.

🔐

Authenticate

The user signs in; the MCP connection carries a scoped credential.

→

🔑

Per-user key

Each user’s MCP key resolves only the brains they are entitled to.

→

👁️

Visibility check

Private, public, open, or organization decides who can query.

KBRAIN

→

✅

Answer

Knowledge is returned only from brains the user may reach.

Access is decided per request. A user only ever sees knowledge from brains their key and the brain’s visibility allow.

Authentication and per-user keys

Every connection is tied to an authenticated account - via email sign-in or an OAuth authorization with a read scope for MCP. Each user gets their own MCP key, and that key resolves only the brains they are entitled to. There is no shared, unscoped connection: access always maps back to a specific user.

Visibility: the four levels

Private: only the owner can query the brain
Public: listed in the marketplace, anyone can subscribe and query
Open: a shareable link that lets specific people query without a full setup
Organization: shared across everyone in the company, reachable through each member’s own key

The organization security model

For companies, access follows organisation membership. An organisation is keyed on verified email domains, so only people with a company address are members. Only owners and Knowledge Managers can publish or share brains org-wide; plain members can query what is shared with them but keep their own brains private. Access to shared knowledge tracks membership automatically.

The default is private. A brain is only reachable by others when its owner deliberately makes it public, shares an open link, or - in a company - shares it org-wide. Nothing is exposed by accident.

Connect your knowledge over MCP

Create or subscribe to a brain on KBrain, then connect its MCP endpoint to Claude, ChatGPT, or any MCP compatible assistant in a few minutes.

Connect via MCP

Frequently asked questions

How does KBrain control who can query a brain?

Through authenticated accounts, a per-user MCP key, and a visibility setting on each brain. The key resolves only the brains a user is entitled to, and the brain’s visibility - private, public, open, or organization - decides who can reach it. The check happens per request.

What are the visibility levels?

Private (owner only), public (listed in the marketplace for anyone to subscribe), open (a shareable link for specific people), and organization (shared across everyone in a company). The default is private, so nothing is exposed unless the owner chooses to share it.

How is access controlled inside a company?

Access follows organisation membership, which is keyed on verified email domains. Only owners and Knowledge Managers can share brains org-wide; members can query what is shared with them and keep their own brains private. Membership - and therefore access - updates automatically.

Is every brain private by default?

Yes. A brain is private unless its owner deliberately makes it public, shares an open link, or shares it org-wide. There is no accidental exposure - sharing is always an explicit action.