Co-authored by Ralf Schmauder and Claudia Zentner

With IBM Cloud Pak for Business Automation 25.0.1 and IBM Business Automation Workflow 25.0.1.0, you can now trigger your workflow automations via a local workflow Model Context Protocol (MCP) server, enabling AI agents to interact with your business processes.

By importing the MCP server into an AI agent platform like IBM® watsonx Orchestrate®, you enable AI agents to dynamically discover and call your workflow automations as contextual tools during conversations. This creates a powerful bridge between conversational AI and your workflow automations. Instead of solely relying on static integrations or predefined triggers, MCP allows agents to select and execute the right workflow automation based on the user’s intent and the context of the interaction.

Using exposed REST services as tools

The MCP server provides operations from exposed REST services as tools to AI agents. REST services exposed from workflow automations are strongly typed. This makes them ideally suited for the MCP server, since they can be easily mapped to well-defined tools that AI agents can invoke.

Importing the MCP server

To import the MCP server, you need access to a workflow environment running version 25.0.1, along with workflow credentials.

The MCP server can be downloaded from https://github.com/ibmbpm/ibm-baw-mcp-server. Detailed instructions for connecting to your workflow environment and importing the MCP server as a toolkit into IBM® watsonx Orchestrate® are provided in the GitHub repository’s README file.

Sample use case: Claim processing

To demonstrate how the MCP server connects conversational AI with your workflow automation, consider the following claim processing example. In workflow, a REST service is exposed that provides four operations: “SubmitClaim”, “InitiateRiskEvaluation”, “CancelClaim”, and “GetClaimStatus”. These operations represent the key steps in handling an insurance claim.

With these new tools in place, an AI agent called “Claims Processing Agent” is created and equipped with those tools. The agent is now ready to interact with customers in a conversational manner. When a customer reports a car accident through the chat interface, the agent asks follow-up questions to gather all necessary details, such as the policy number, the vehicle model, and a description of the incident. Once the required information is collected, the agent invokes its “SubmitClaim” tool, which triggers the corresponding REST operation in the workflow system.

Finally, the workflow end user interface shows that a new user task has been created for an insurance specialist to review. This task contains all the information gathered during the conversation, ensuring that the claim process continues seamlessly from the AI-driven interaction to the human review.

This example illustrates how the MCP server enables AI agents to dynamically discover and invoke workflow automations based on user intent.

Conclusion

The workflow MCP server provides a seamless way to integrate agentic AI with your existing workflow automations. It acts as a bridge between AI agents and your enterprise system, enabling natural language interfaces without additional integration effort. This approach combines the reliability of established processes with the flexibility of AI-driven interactions, unlocking new possibilities for innovation while maintaining predictability.

References

• IBM Cloud Pak for Business Automation 25.0.1 documentation
• IBM Business Automation Workflow 25.0.1.0 documentation
• Local workflow MCP server GitHub repository
• IBM® watsonx Orchestrate® MCP servers documentation