AWS exposes Step Functions workflows as AI agent tools
AWS announced on October 29 that AI agents can use AWS Step Functions workflows as tools through Amazon Bedrock AgentCore Gateway. The connector supports both Standard and Express workflows and uses the Model Context Protocol to connect an agent to the configured tools.
The agent decides when to invoke a workflow and supplies its input. Step Functions controls the steps and their sequence. That separation can keep a multi-step business process in a defined order, but it does not by itself prove that the workflow encodes the right business policy or that every input is valid.
AWS gives refunds as an example: a configured workflow can run an eligibility check, manager approval and credit step in sequence. The announcement also describes looking up workflow runs to identify failed steps. These are capabilities and examples, not evidence that every customer has deployed them.
The developer tutorial says the gateway creates tool schemas and calls Step Functions without a custom Lambda wrapper. Its example configures six operations, including starting, listing, inspecting and stopping executions. The full connector has ten operations; the tutorial notes that listing executions, describing executions and retrieving execution history do not work with Express executions.
Permissions remain part of the setup. Start and list tools are pinned to the selected state machine, while tools acting on an individual execution accept an execution identifier. The gateway service role’s IAM policy limits which executions those tools can access. Teams still need to configure that role and the exposed operations for their intended use.
Availability follows the AWS Regions where Amazon Bedrock AgentCore Gateway is offered, not a blanket promise for every commercial region. AWS says standard Step Functions and AgentCore Gateway rates apply. The tutorial warns that resources can continue to incur charges until removed, so this announcement should not be described as a free integration.