Administering AI settings

AI settings provide a centralized interface for enabling AI features, managing IBM watsonx® models, registering Model Context Protocol (MCP) servers with agents, and controlling multi‑agent execution. By centralizing AI configuration in a single location, AI settings help ensure consistent behavior across the platform, supports governance requirements, and reduces the risk of unintended or uncontrolled AI usage.

About this task

Core areas of AI settings
AI settings are structured around the following core areas, each of which addresses a specific aspect of AI governance and execution.
  • AI enablement and governance - AI capabilities are available on an explicit opt‑in basis, and administrators must review and accept all applicable AI usage and model‑specific licensing terms to enable AI functionality. This approach helps ensure that AI features are enabled intentionally and in alignment with organizational policies. By requiring explicit acceptance of terms, AI settings support transparency, compliance, and responsible adoption of AI features across environments.
  • External tool integration - AI agents cannot directly call external systems or tools on their own. An MCP server provides a controlled and governed way for agents to interact with external services. By adding an MCP server, you define:
    • Which external system an agent can access
    • What actions (tool operations) the agent is allowed to perform
    • How access is authenticated and secured

    MCP servers are associated with specific agents, helping ensure that each agent can use only the external tools that are explicitly configured for it. Administrators register and configure MCP servers and explicitly enable the tool operations that agents are allowed to use. AI agents use only the tool operations that administrators make available, helping ensure controlled, and auditable access to external systems.

  • Agent execution and orchestration - AI agents can run tasks independently or participate in coordinated execution through multi‑agent mode. In multi‑agent mode, the Coordinator orchestrates task execution across multiple agents, enabling collaboration and sequencing when required. This orchestration capability affects execution behavior only. Changing execution mode does not delete agents or modify their configurations, allowing administrators to adjust execution behavior without impacting agent definitions.
Access and role‑based control
Access to AI settings is restricted to users with specific administrative roles. Role‑based access helps ensure that only authorized users can enable AI features and control agent behavior. The following roles can view and use AI settings. Only users with one of these roles can enable or disable AI features, manage MCP servers, control multi‑agent execution, and enable or disable tool operations.
  • iPaaS admin - A subscription‑scoped role that allows management of AI settings across the entire subscription.
  • Service admin - A service‑scoped role that allows management of AI settings only for the specific service to which the role is assigned.
AI agents and configuration status
AI agents perform AI‑assisted tasks by using AI models and available MCP tool operations. If AI access is disabled, agents and tools remain unavailable until access is re-enabled. AI settings provide visibility into the overall AI configuration, including agent status, MCP server connectivity, enabled tool operations, last updated timestamps, and availability of AI capabilities.
Usage considerations
When administering AI settings, keep the following considerations in mind.
  • MCP servers must be configured and managed through the AI settings page.
  • AI agents can use only the tool operations that administrators explicitly enable.

Procedure

  1. To access AI settings, sign in to webMethods Hybrid Integration and from the left navigation panel, select Administration > AI settings. The AI settings Overview page is displayed.
  2. To enable AI features, follow these steps.
    1. In the Overview section, turn on the watsonx.ai® service toggle.
    2. Review the list of available AI models.
    3. Select View model details to review the AI model information.
    4. Select Read terms to review the licensing information.
    5. Select I confirm that I have reviewed the model details and agree to the AI usage terms.
    6. Select Agree and continue.
      AI features are now enabled across webMethods Hybrid Integration and the models are now available for agent execution.
  3. Multi‑agent mode enables coordinated execution of tasks across multiple AI agents by using a single orchestrator (Coordinator). When multi‑agent mode is enabled, the Coordinator coordinates task execution across agents. Multi‑agent mode affects execution behavior only and does not delete agents or configuration data. To enable multi‑agent execution, follow these steps.
    1. Go to AI settings > Overview.
    2. Set the Multi‑agent mode toggle to On.
  4. MCP servers allow AI agents to use external tools through a standardized protocol. To add an MCP server, follow these steps.
    1. In AI settings > Agents, select the agent for which you are registering the MCP server.
    2. Select Add to begin registering a new MCP server.
    3. In the Name field, enter a name for the MCP server.
    4. In the Connection URL field, enter the endpoint URL of the external service. The connection URL identifies the external service endpoint that the MCP server represents.
    5. Select Next. In the Authorization type field, select Token. In the Token field, enter the authentication token that is required by the external service. Use token as the authorization type when the external service requires a token‑based credential to authenticate requests from the MCP server. With token authorization, the MCP server includes a security token with each request it sends to the external service. The token authenticates and authorizes the request, allowing AI agents to securely invoke approved tool operations without exposing credentials in agent logic.
    6. Select Test Connection to verify that the MCP server can connect to the external service using the provided URL and token. Confirm that the connection test succeeds.
    7. Select Next and review the tool operations exposed by the MCP server.
    8. Enable the required operations. An MCP server exposes multiple tool operations, but not all are available to agents. Explicitly enabling operations allows administrators to restrict agents to approved actions. Only enabled operations are discoverable and usable by agents.
    9. Select Add. The MCP server is registered and available for the selected agent. Only enabled tool operations are available for agent execution.
  5. To enable or disable MCP tool operations, follow these steps.
    1. In AI settings > Agents, select an agent.
    2. Select an MCP server and then select one or more tool operations.
    3. Select Turn on to enable the tool operation, or select Turn off to disable it.