Optional AI connections

This notice describes the host-neutral SafeGora MCP connector operated by OllieLabs LLC. It supplements the SafeGora Privacy Policy when you choose to connect an existing account through a separately enabled AI host. It does not imply that every host is enabled or approved for directory publication. Accounts remain subject to the Terms of Service, including the minimum age of 18.

Information used

SafeGora uses account, workspace and identity-provider identifiers, current membership and role, application consent and a verified native sign-in claim to check access. WorkOS handles secure sign-in and account linking. Authentication secrets belong in the sign-in interface, not in conversation prompts or tool inputs.

Permitted tool results include opaque person/workspace identifiers, workflow and source-record identifiers, operational statuses, dates, counts, advisory readiness or loss-control values, proposal status and native continuation links. Proposal inputs contain bounded record references, dates, priorities and a request identifier needed to prepare and retry a draft.

Names, locations, uploaded narratives and documents, medical and injury records, government identifiers and payment-card data are excluded from these tools. SafeGora does not receive your full conversation through the connector. Human verification, certification, filing and communications remain in SafeGora.

Purpose and recipients

SafeGora uses the information to authenticate the connection, enforce current workspace access, answer requested workflow lookups and prepare proposals for native human review. Authorized native reviewers can see proposals and required audit evidence under existing access rules.

SafeGora does not use connector inputs or results for model training, advertising or profiling. The connector returns the permitted native results to the requesting host; it does not send them to SafeGora's own AI provider.

The host you select receives the permitted results of your requests: Anthropic for Claude, Meta for Muse, or xAI and the applicable Grok service providers for Grok. Where Grok Bot uses Cursor infrastructure, that infrastructure also processes the connection. SafeGora does not send a result to the other listed hosts merely because their integrations exist.

The receiving host's privacy policy and your account settings govern its handling of data. WorkOS and SafeGora's existing hosting and security providers support the service as described in the sub-processor list. The separate OpenAI integration is described in the ChatGPT connection privacy notice.

Verification and retention

A pending verification challenge is valid for ten minutes. SafeGora stores its hash. Your browser tab temporarily holds the challenge during sign-in and removes it when the native verification page consumes it or the tab closes. A challenge alone grants no record access.

When a native disconnect completes, SafeGora immediately deletes this connection's saved preferences and removes its challenge hash, identity-provider user and organization binding, native-session reference and saved permission mode from the revoked binding. This applies to this person, host and current workspace. You can use disconnect to request removal of those connector-only fields.

How long information is kept

• Connection binding. Removed immediately when you disconnect in SafeGora, as described above.
• Revoked-consent marker. After disconnect, SafeGora keeps only a revoked status, the consent identifier, opaque person and workspace identifiers, the host client and endpoint, and timestamps. This stops an old authorization from reopening access. The marker is kept while the workspace account exists and is handled with the account's other data when the account ends, as set out in the Terms of Service.
• Proposals, run records and audit evidence. Records created through the connector become native records in your workspace, kept and controlled like other workspace records for the life of the account. Audit evidence is append-only and hash-chained, so individual entries are not edited or deleted separately. Disconnecting stops host access to these records.
• Operational and security logs. Kept for the periods in section 6 of the SafeGora Privacy Policy.

Removing access only in the host stops subsequent access checks, but does not clear the saved connection fields in SafeGora. Disconnect in SafeGora as well, or ask SafeGora to remove them.

To request access to or deletion of connector-related information, follow section 7 of the SafeGora Privacy Policy or email Ollie@safegora.com. Requests remain subject to your organization's instructions as the workspace customer, legal holds and recordkeeping obligations that apply to the workspace.

Your controls

Choose Read only on the native verification page to disable proposal preparation, new connection challenges and audited lookups after connection. Read and write access includes those tools, with native human review still required for proposals.

In SafeGora, open Settings, Integrations, then Manage other AI connections. Choose the host and disconnect your current workspace. Each host has a separate connection. Removing one does not revoke your other hosts or change another person's access. You can also remove the OAuth connection in the selected host.

Revocation stops subsequent workflow access. It does not erase information already returned to the host or stored in your conversation history; use that host's controls for those records.

For other questions, contact SafeGora. Technical setup and tool boundaries are covered by the MCP connector guide.