AI assistants (MCP)
Permissions and limits
What an assistant is allowed to do, and what stops it.
Permissions
| Permission | Allows |
|---|---|
projects:read |
List and read projects. |
cases:read |
Search and read cases, flows and milestones. |
cases:write |
Create and update cases, flows and milestones. Never delete. |
manual-runs:read |
Read manual runs and results. |
manual-runs:write |
Start runs and record results. |
Three things must agree
A call is allowed only if all three permit it:
- the permissions on the token or connection;
- the projects chosen when it was made;
- that member's access right now.
Remove somebody from a project and their assistant loses it on the next call. Remove them from the organization and every token they made stops.
Also enforced
- Business plan, checked on every call.
- Two-factor, if your organization requires it: a member who has not enrolled cannot use a token.
- Suspended organizations are refused.
- 120 calls a minute per token.
What is recorded
Every change an assistant makes is written to the audit log as that member, naming the token: "[email protected] via token "Claude Code": created the test case "Refund a paid invoice"". Creating and revoking a token, and connecting an app, are recorded too.
Where to look
Profile → API tokens for your own. Organization → Settings → Security for everybody's, where an owner can revoke any of them.