Relay Workflow Automation MCP. Manage complex process runs and check status via AI agent.

You're connecting your AI client straight into Relay's no-code process platform. This MCP Server lets you treat entire business operations as if they were a function call—you manage the full lifecycle of complex automations, not just simple data points. You gotta run whole pipelines here.

Discovering What’s Available

To see what's running in your account, use list_workflows. This tool dumps every single automation workflow configured in Relay. It gives you a list of all the available processes, so you know exactly which ones you can trigger or check on. If you need more details about one specific process—like its input requirements or internal logic—you run get_workflow and pass it the unique workflow ID to retrieve that detailed metadata.

Kicking Off a Workflow Run

The core action here is triggering a job. You use run_workflow to start an automated business process. When you call this tool, you gotta provide specific input variables structured as JSON data. These inputs are required by the workflow—think customer email addresses, invoice IDs, or specific product SKUs—and they fuel the entire run.

Once that payload hits Relay, a new execution job starts immediately.

Tracking and Controlling Jobs

Once you kick off a run, you need to know what's happening. You call get_run_status, passing in the unique run ID from the initial execution. This tool tells you the current state: is it running right now? Did it fail already? Or did it finish up clean?

If you need to see a history of these runs—the good, the bad, and the canceled—you use list_runs. This gives you a list of recent run IDs along with basic status data, letting you quickly check on past jobs. And if something goes sideways, if that job just hangs in purgatory, you can hit the brakes.

You use cancel_run and pass it the specific run ID to forcefully halt an automation that's still kicking.

Operational Flow Example

Imagine your agent needs to process a batch of customer records. First, it calls list_workflows just to confirm the 'Customer Onboarding' process exists. Next, it uses get_workflow on the returned ID to validate that the workflow expects an array of email addresses in its JSON payload. It then constructs that payload and fires off the run using run_workflow.

The moment the job is created, it captures the new Run ID. With that ID, your agent immediately calls get_run_status every minute until the status flips to 'completed' or 'failed.' If, after ten minutes, the status still reads 'running,' you know what to do: grab that same run ID and execute cancel_run.

You manage the whole sequence—discovery, execution, monitoring, and termination—all through these direct tool calls. It’s a full-stack operational control layer for your business processes.