Administer MCP

MCP is off by default. Administrators enable the feature, register approved clients, choose the maximum data scopes, maintain organization-wide exclusions, and review activity.

Enable MCP

  1. Open Configuration Hub > Feature Switches.
  2. Find MCP.
  3. Review your organization's AI and data-sharing policy.
  4. Turn on the feature.
  5. Confirm that MCP pages appear in the Integration Hub.

If registration reports that the MCP resource identifier is not configured, the deployment is not ready for client registration. Contact the system operator; this cannot be fixed from the client page.

Register a client

Open the external client's OAuth setup beside RiskVault so you can copy the exact redirect URI and determine whether it supports a secret.

  1. Open Administration > Integrations > MCP Clients.
  2. Under Register a new client, enter a friendly display name.
  3. Enter a unique client ID. It cannot be changed later.
  4. Choose Confidential if the client asks for both an ID and secret. Choose Public only when it cannot safely store a secret.
  5. Copy the client's exact HTTPS redirect URI into RiskVault.
  6. Select at least one scope:
  7. Risk summaries allows search and summarized risk details.
  8. Portfolio-wide statistics allows approved aggregate metrics.
  9. Leave proof-of-possession required. Select the DPoP exemption only after confirming that the client cannot support it.
  10. Select Register client.
  11. For a confidential client, copy the displayed secret immediately through an approved secure channel. RiskVault does not show it again.
  12. Enter the client ID, secret if applicable, RiskVault authorization information, and MCP server URL in the external client.

Image needed: MCP Clients registration form with Display name, Client ID, type, Redirect URI, scopes, DPoP option, and Register client labeled.

Warning

A DPoP exemption weakens token theft protection. Document the compatibility reason and revisit it when the client gains support.

Edit or suspend a client

From the registered-clients table you can:

  • edit its display name, redirect URI, scopes, and DPoP setting;
  • disable it to refuse new authorization without deleting the registration;
  • roll a confidential-client secret when it approaches its recommended expiry;
  • permanently delete it when it should never connect again.

Client ID and type cannot be changed. Register a replacement instead. Built-in infrastructure identities are managed automatically.

Set organization-wide data exclusions

  1. Open MCP Clients > Data sharing.
  2. Review business units and risk categories.
  3. Exclude any area that no external AI tool should receive.
  4. Save and test with a user who would normally see that area.

These exclusions apply to every MCP client and user. They sit on top of the fixed classification restriction: MCP only returns Public and Internal data.

Review activity

Open Administration > Integrations > MCP Activity. The page shows a rolling 30-day summary of tool calls, denials, distinct callers, and classifications, plus a filterable event list.

Review activity when:

  • a client is first introduced;
  • usage changes unexpectedly;
  • a user reports a denied request;
  • a secret or token may be exposed;
  • a client is being renewed or retired.

Image needed: MCP Activity page with the 30-day summary cards and event filters, using fictional activity.

Respond to a suspected compromise

  1. Disable the affected client immediately.
  2. Review MCP Activity for the client, user, time window, tools, and denials.
  3. If only a confidential client secret may be exposed, roll it and update the client securely.
  4. If trust cannot be restored, delete the client and register a replacement with a new ID.
  5. Review data-sharing exclusions and affected users' permissions.
  6. Follow your organization's incident-response process.