How Klarify AI applies your permissions

When you connect an AI client to Klarify, the client inherits your Klarify role — it can do only what you can do in the app. For the underlying role model, see Permissions and roles.

Tool visibility by role

The MCP server only exposes the tools your role is allowed to call — lower-privileged users do not see administrative tools at all.

Tool group	Guest	Employee	Org Admin	Super Admin	Account Manager	Account Owner
Session	✅	✅	✅	✅	✅	✅
Organization info	✅	✅	✅	✅	✅	✅
Organization stats	—	✅	✅	✅	✅	✅
Organization settings	—	—	✅	✅	✅	✅
Organization billing	—	—	—	—	—	✅
Read tools (employees, departments, documents, etc.)	—	✅	✅	✅	✅	✅
Write tools (create / update / delete)	—	—	✅	✅	✅	✅

How role enforcement works

Role checks happen on the MCP server, not in the AI client. Even if a tool name appears in an AI client’s tool list, the server rejects calls that exceed the user’s role.

For example, an Employee who asks the AI to “delete the Sales department” will get a refusal — the delete_department tool is not registered for that session.

How write actions are approved

When you ask the AI to make a change — create a position, update an employee, delete a team — the change is not applied immediately. Each write is intercepted and shown to you as an approval card describing the exact action and its values. The change runs only after you approve it; approving the card is your consent for the specific values shown.

A few behaviors follow from this:

One action at a time. If you ask for several changes in a single message (“create a department and a team”), the AI proposes them one approval card at a time rather than all at once.
Deletions that would orphan data ask first. Before deleting an employee who holds positions, or a position that owns documents, the AI asks how to reassign that work — to another employee or position, or to leave it vacant — and only then proposes the deletion.
Read-only sessions cannot write. Users without write permission cannot make changes through the AI at all; the assistant declines change requests and suggests contacting an organization admin.

Content-level access

Content-level permissions apply through the AI client just as they do in the app. Folders are visible to every active member, but individual documents inside a folder can be restricted. For example, an Org Admin who has not been granted access to a specific document cannot read that document through the AI client. See Permissions and roles for the underlying model.

Some content tools depend on your content access role rather than your admin role. The document-visibility tools, for instance, are available to members with Editor or Manager content access and only act on documents you can edit — so a member with no admin role can still use them, while a Viewer cannot, regardless of admin role.