Para agentes de IA: um índice de documentação está disponível em https://www.mongodb.com/pt-br/docs/llms.txt — as versões de markdown de todas as páginas estão disponíveis anexando .md a qualquer caminho de URL.
Menu Docs

Manage AI Client Access to Your Organization

As an Organization Owner, you control whether AI clients can connect to your Atlas organization through user-delegated access, and what level of access those clients receive. AI clients can't connect to your organization until you enable AI client access.

Atlas disables AI client access by default for existing organizations and for new organizations an existing Atlas user creates. Atlas enables AI client access by default for organizations new users create at the time they create their Atlas account.

When AI client access is enabled, an AI client must be authorized by an individual Atlas user to act on that user's behalf. An AI client acts using its authorizing Atlas user's existing Atlas permissions and can't exceed them. A user's ability to read or modify cluster data is governed by that user's Atlas role and your organization's access mode. Before you enable AI client access, review which roles can connect AI clients to your clusters in User Roles and Database Access.

You can manage AI client access settings in the Atlas UI.

When AI client access is enabled, users who hold any of the following Atlas roles can authorize an AI client to connect to your organization's clusters:

  • ORG_OWNER

  • GROUP_OWNER

  • GROUP_DATA_ACCESS_ADMIN

  • GROUP_DATA_ACCESS_READ_WRITE

  • GROUP_DATA_ACCESS_READ_ONLY

  • GROUP_STREAM_PROCESSING_OWNER

An AI client's access to cluster data is the more restrictive of two limits: the authorizing user's Atlas role and your organization's AI client access mode. The access mode can only reduce a user's access, never grant more than the user's role allows. For example, a user with the GROUP_DATA_ACCESS_READ_ONLY role has read-only access through an AI client, even when the organization allows AI clients to read and write.

All tool calls are attributed to the individual user who authorized the AI client that made the call. You can identify which user performed each action in your audit log.

Atlas records new audit events for App Connections and adds delegation context to existing audit events, so you can distinguish actions taken through an AI client from other activity.

To learn how Atlas attributes each action to an individual user, see Security, Governance, and Auditability.

Observação

Read-only Atlas Administration API calls are not recorded in the audit log. This is an existing Atlas behavior and is not specific to AI client access.

To enable AI client access for your organization, complete the following steps:

1

In the Atlas UI, open your organization settings, then select App Connections.

2

Choose the access mode that AI clients receive when they connect. For a description of each mode, see AI Client Access Modes.

3

After you save, AI clients can connect to your organization using the access mode that you selected.

AI client access supports the following modes:

  • Read: AI clients can run read operations, including find, aggregate, count, list-collections, list-indexes, and schema inspection.

  • Read and write: AI clients can run read operations and write operations, including insert-many, update-many, create-collection, and create-index.

  • Disabled: No AI client can connect to your organization.

Tools outside the configured mode are not registered with the AI client's session and are not visible to the client. For example, in read mode, write tools do not appear in the client's list of available tools.

To stop all AI client access to your organization, set the access mode to disabled. This prevents any new connections and immediately ends existing control-plane access for every AI client connected to the organization. Data-plane access through artifacts that a client already created, such as database users, continues to work.

Individual users can revoke their own connections at any time from the Atlas UI.

For revocation timing and how to handle federated users, see the revocation and lifecycle guidance in atlas-app-connections.

By default, an AI client's access on behalf of a user ends after 7 days of inactivity, or 30 days from when access was initially granted regardless of activity. When this access ends, the user must reauthenticate through the AI client's plugin to reconnect the AI client.

To learn how AI clients connect to your database using short-lived credentials, see Security, Governance, and Auditability.

As an Organization Owner, you can reduce the maximum token lifetime for your organization. Reducing the maximum lifetime shortens the window during which a connection remains active without reauthentication.

If you rely on an identity provider (IdP) to manage Atlas access, reduce the maximum token lifetime so that access changes you make in your IdP take effect sooner. For example, if you revoke a user's access in your IdP, a shorter token lifetime reduces how long an AI client can continue acting on that user's behalf before it must reauthenticate.

AI client access applies at the organization level. You cannot set per-user, per-cluster, or per-client access controls, and the access mode applies to the entire organization.

AI client access also shares the limitations of Atlas App Connections, including but not limited to the inability to:

  • Restrict access to specific AI clients

  • Revoke an individual user's connection as an Organization Owner

  • Delegate a subset of a user's permissions

For the complete list of limitations, see Limitations.

For the delegated-access concepts that underlie AI client access, including authorization, consent, and lifecycle, see atlas-app-connections.