Administer AI

Use the AI Admin console to:

Prerequisites

To access the AI Admin console, you need the AI Administrator privilege.

Set the Retention Period for Agents and Explorer Chats

Set this to comply with your organization's data retention policies and manage metadata and storage service growth. You have greater control over AI-generated data lifecycle management. An agent's author can override the retention period for the specific agent.

  1. In the Library sidebar, click AI Admin.

  2. In the AI Admin sidebar, click General Settings.

  3. By default, these conversations are retained indefinitely. You can select a different retention period:

    • Retain History for 30 Days

    • Retain History for 90 Days

    • A specific period: Select Retain History and type the number of days.

      If you define a retention period defined, when the conversation history expires, it is automatically removed when the agent is opened or the user logs into Library. This includes the associated snapshots.

  4. To return to Library, click the Library logo in the sidebar.

Define Access to Explorer Tools

  1. In the Library sidebar, click AI Admin.

  2. In the AI Admin sidebar, click Tools Access.

  3. By default, sharing via email is disabled. To enable it, toggle on Share Via Email. Configure the email address and smart host using Explorer.

  4. By default, HTML creation is disabled. To enable it, toggle on HTML Creation.

  5. By default, web search is disabled. To enable it, toggle on Web Search.

  6. When Web Search is enabled, you can whitelist and blacklist domains:

    • To specify the domains that can be used in web search, click Only use these domains in web search. Enter each URL to allow.

    • To specify the domains that cannot used in web search, click Never use these domains in web search. Enter each URL to blacklist.

    • You can edit and delete blacklisted and whitelisted URLS. Hover your cursor over the URL and click Edit or Delete.

  7. To return to Library, click the Library logo in the sidebar.

Add and Connect to an MCP Connection

Before you begin:

  • Obtain the MCP server URL and authentication details from the MCP provider.

    • For API Key authentication, obtain a credential from the MCP provider. You need the information listed here.

    • For OAuth authentication, create or obtain an OAuth application from the MCP provider. Register the Library MCP callback URL for that OAuth application. If you do not know the callback URL for your environment, contact your Library administrator. You need the information listed here.

    • If the provider changes its OAuth application settings, client secret, or permitted scopes, update the connector and connect again.

  • Ensure that the server supports MCP over HTTP.

  1. In the Library sidebar, click AI Admin.

  2. In the AI Admin sidebar, click MCP Connections.

  3. Click the Add icon next to Connectors.

  4. Fill in the following information:

    Parameter Required/Optional Description
    Name Required

    A unique name for the connector.

    Use English letters and numbers only, without spaces or special characters.

    Example: GitHubMCP
    Server URL Required

    The complete URL of the MCP endpoint, not the provider's website or documentation page.

    Example: https://api.githubcopilot.com/mcp

    Description Optional A short explanation of what the connector provides.
    Logo URL Optional A publicly accessible HTTPS URL for an image that represents the connector.

  5. Choose an Authentication Mode from the drop-down list:

    • If the MCP server explicitly allows requests without authentication, select None. The MCP server could be intentionally public or otherwise accessible without credentials.

    • If the provider gives you a static credential, select API Key. The static credential may be called an API key, personal access token, or other static token. Fill in the following information:

      Parameter Required/Optional Description
      Header Name Required

      The HTTP header used to send the credential. Common values are Authorization and X-API-Key. Use the exact name specified by the MCP provider.
      Header Value Template Required

      The format used to place the API key in the header. The template must contain the {apiKey} exactly as shown.

      Examples:

      • If the provider requires Authorization: Bearer <token>, the Header Name is Authorization and the Header Value Template is Bearer {apiKey}.

      • If the provider requires Authorization: token <token>, the Header Name is Authorization and the Header Value Template is token {apiKey}.

      • If the provider requires X-API-Key: <key>, the Header Name is X-API-Key and the Header Value Template is {apiKey}.
      API Key Required

      The secret credential issued by the MCP provider.

      Treat this as a password. Do not include it in the connector name, description, Server URL, Logo URL, or support screenshots.

    • If the provider requires each user to sign in and grant access through OAuth, select OAuth. Fill in the following information:

      Parameter Required/Optional Description
      Authorization Endpoint Required The provider URL where users sign in and approve access. This is normally an endpoint ending in a path such as /authorize, not a general login page.
      Token Endpoint Required The provider URL used to exchange the authorization result for an access token.
      Client ID Required The public identifier of the OAuth application registered with the provider.
      Client Secret Required The confidential secret of the OAuth application. Copy it from the provider's application settings.
      Scopes Optional

      The permissions requested from the provider. Enter only the scopes required by the MCP tools you plan to use. Follow the format shown by the provider; in this form, multiple scopes can be entered as a comma-separated list.

      Use the least-privileged scopes that allow the required MCP tools to work.

  6. Click Add.

  7. Click Connect.

    • If the authentication mode is set to None, no authentication header is sent to the MCP server. If the server returns an authentication error, edit the connector and select API Key or OAuth instead.

    • If the authentication mode is set to API Key, the information is sent to the provider and you are connected without manually signing in.

    • If the authentication mode is set to OAuth, sign in on the provider's authorization page. Review the requested permissions and approve access.

  8. Message or Symptom Fix
    The server requires authentication Change the Auth Mode from None to API Key or OAuth.
    The API key was rejected Verify the credential, header name, value template, expiration date, and provider permissions.
    The server returns HTTP 403 or reports insufficient scope Verify that the Scopes field contains the exact permissions required by the MCP provider and uses the expected format. Also confirm that the OAuth application is allowed to request those scopes, then connect again and approve the updated permissions.
    OAuth authorization is invalid Connect again and complete the authorization flow. Verify the client ID, client secret, endpoints, callback URL, and scopes.
    The MCP endpoint is invalid Verify that the Server URL points to the MCP endpoint rather than the provider's website.
    The server cannot be reached Check the URL, network access, proxy, firewall, DNS, and TLS certificate.

Edit an MCP Connection

  1. In the Library sidebar, click AI Admin.

  2. In the AI Admin sidebar, click MCP Connections.

  3. Select the MCP connection.

  4. In the toolbar in the upper right, click the Edit icon .

  5. See Add and Connect to an MCP Connection for detailed information on each field.

  6. Save your changes.

Disable or Enable an MCP Connection

  1. In the Library sidebar, click AI Admin.

  2. In the AI Admin sidebar, click MCP Connections.

  3. Select the connection in the Available list and click Disable.

  4. To enable a disabled connection, select the connection in the Disabled list and click Enable.

Delete an MCP Connection

  1. In the Library sidebar, click AI Admin.

  2. In the AI Admin sidebar, click MCP Connections.

  3. Select the MCP connection.

  4. In the toolbar in the upper right, click the Delete icon .