Governing connected assistants

Two levels of control exist. Users manage their own connections; the rest sits with Modero and reaches you through support.

What a user controls#

Profile > Connected Applications lists every AI client connected to that person’s account, with Revoke next to each one. Revoking immediately drops all of that client’s tokens for that account, and it cannot act as them again until they re-authorize.

A single connector often registers many times behind the scenes, so the list collapses them into one row per application. One Revoke disconnects the whole thing.

Anyone can do this without an administrator, and it is the reliable way to cut access. Removing the connector inside the AI client stops that client from using the connection but leaves the authorization standing, so do both when removing an assistant for good.

Changing a password, disabling an account, or removing someone from an organization all end their connections as well.

What you control#

Your levers are the entitlements described in Turning MCP on:

• Withhold api_write on a client’s plan. Reads work, every write is refused at the platform. This is the right lever for “let them look, not touch” across a whole organization.
• Withhold mcp_access entirely, and no assistant reaches that organization’s data at all.
• A user’s role. Because a connection runs as its user, narrowing the role narrows the assistant immediately, on the next request. There is nothing separate to revoke.

What Modero controls#

The platform governance console holds the client-level levers. Ask support when you need one of these:

• Approve, deny, or revoke a client. Needed for any AI client outside Claude, ChatGPT, and Gemini.
• Force read-only. A client flagged read-only cannot write, no matter what the user’s role is. Use this when the constraint is about the tool rather than the plan.
• Assign user. Pin a client to operate as a specific user. Its reach is then computed from that user’s live role and memberships, and it can never grant more than that user already has.
• Revoke one registration. A single rogue registration can be revoked surgically rather than dropping the whole application.

Revoking takes effect mid-session. A revoked client’s next tool call is refused, not just its next sign-in.

What is never exposed#

Creating or editing users, organizations, or partners is not available over MCP at all. Those actions stay in the Partner Hub, for anyone, at any tier. There is no flag that turns them on.

Handling a “revoke this now” request#

1. Have the user open Profile > Connected Applications and revoke it. That is immediate and needs nobody else.
2. If they cannot reach their profile, disabling the account ends every connection it holds.
3. If the concern is the tool rather than the person, escalate to Modero for a client-level revoke or a force read-only flag, and say which client and which organization.
4. Check the audit feed for what the assistant actually did. MCP-originated changes are distinguishable from web-app ones.

Common questions#

Can a Modero platform admin read a client’s data through their own AI assistant? A platform admin’s reach over MCP is unbounded for reads, matching their authority in the app. Partner-plane and organization-plane concealment rules, including support-access grants and legal hold, continue to apply wherever they apply in the app. See Modero support access.

A user revoked a connector and it still works. They most likely removed it in the AI client rather than revoking it under Connected Applications. Removing it client-side stops that client from calling, but leaves the authorization standing.