Coming in IBM Robotic Process Automation (RPA) 30.0.2 is the ability to invoke RPA bots via the official model context protocol (MCP) server to enable RPA bot integration with your AI agents. This blog is a step-by-step guide for building your first MCP-enabled RPA bot.

Building the bot

For the purpose of the demo we'll build a very simple RPA bot. It will take an input parameter and return it concatenated to another string which demonstrates data flowing between your AI agent and the bot.

IBM RPA bots are built in the Studio application, documentation for using the Studio can be found here.

It is important to give any input/output variables sensible names that can help an AI agent contextualize their purpose and adding comments to the variables will further help the AI agent understand their purpose.

This is my WAL script for the bot, the logMessage command is optional but will help verify all is working correctly in the IBM RPA Control Center.

defVar --name input_name --type String --value Tester --parameter  --required  --comment "The name of the person to greet"
defVar --name hello_output --type String --output  --comment "The returned greeting message"
concatTexts --text "Hello, " --value "${input_name}!" hello_output=value
logMessage --message "${hello_output}" --type "Info"

Once you've built a bot in the IBM RPA Studio, publish it and mark it as production and schedulable.

Configuring IBM RPA

Login to the IBM RPA Control Center to begin configuring your bot to be MCP-enabled.

Creating a Computer Group

To run unattended IBM RPA bots you will need to configure a group of computers which will run the script. If you have already configured a Computer Group, you can skip this step.

Navigate to Computers > Computer groups and create a computer group with at least one available computer.

Creating a Project

An IBM RPA Project is a collection of one or more bots, the project name and description are exposed to the AI agent, so it is recommended that they are related to the containing bots to help the agent understand the contained bots purposes. To create a Project, navigate to the Projects tab and click the Create project button. Make sure your Project has the correct shared access for the user which will be running the bots via MCP. In this blog I will use the "Everyone" group, but you may wish to create a new "MCP Users" group for this purpose.

Once created, navigate to your Project and click the Create bot button

The first page will ask you to give the bot a name and description, make sure these define the purpose of your bot as this will help the AI agent understand when to call the bot. To expose this bot to the MCP server, make sure the MCP tickbox is selected.

On the next page, select the script you created and the Computer Group to run the script.

Review the summary on the last page, if all looks good then create the bot!

Testing your bot (optional)

At this stage I like to run the bot directly from the IBM RPA Control Center to verify everything is configured correctly. Select the vertical ellipses (...) on your RPA bot and select Run.

Enter your input variable and click Run!

Navigate to the Bot runs tab to see your bot running:

Refresh the list after a couple minutes and verify your bot has successfully finished. Once finished, click on your bot and navigate to the Logs tab to verify the log we added to the script looks as expected.

Enabling call-backs

Depending on your MCP serer setup, you may need to enable call-backs in your RPA tenant. If you are using the IBM-hosted SaaS MCP server then you will need to enable this. If you are self-hosting the MCP server on-premises then speak with your SRE team to understand if you should enable call-backs.

IBM RPA's call-back mechanism will return the result of a bot run asynchronously to the MCP server (if the MCP server if configured for this behaviour). This requires that the call-back address be whitelisted from your RPA tenant. This can be done as the tenant administrator by navigating to Tenants > Security settings and creating a new entry for *.rpa.ibm.com which is the IBM-hosted SaaS MCP server, if you are using an on-premises MCP server then this should point to wherever you're hosting it.

Obtaining an IBM RPA API key

Navigate to your profile on the Control Center by clicking the top-right person icon and navigating to Profile.

Under the Identity tokens section click Create and generate an IBM RPA API key, make sure you save this somewhere secure for later.

Connecting to the RPA MCP server

You're all set to connect to the RPA MCP Server!

Connecting your AI agent

In your AI agent of choice, configure the following:

Transport: The MCP server uses the Streamable HTTP transport protocol.

URL: The URL of the MCP server, this should end with the path /mcp.

Headers: Create a header with the name x-api-key and set the value to your IBM RPA API key.

Timeout: Set a large timeout to ensure your AI agent waits long enough for MCP tools.

Due to the long-running nature of bots, you will often need to increase the tool timeout in your MCP client to enable bots to run to completion. If this is not configurable in your MCP client, you may set the flag mcp_wait_for_bot to false when invoking an RPA bot. This is a parameter injected into all RPA bot tools which determines if the MCP server will wait for the bot to respond, or just invoke the bot and return. If you chose to not wait for the bot to complete, you can check on the status (and output) at a later point by using the get_rpa_bot_status tool exposed in the MCP server.

Hello, World!

Summary

In this blog we've written an IBM RPA bot, exposed it to our MCP server, and used Claude to invoke the bot. While the bot created in this blog is quite simple, you can take everything learned and build a more complex bot which can drive your business processes all from your AI agent of choice.