On Demand
Starts the automation when a person runs it, or when another automation calls it as a step. Reach for it for anything you kick off yourself, and for a reusable sub-automation that other automations invoke through the Automation action.
Fields
| Field | Description |
|---|---|
| Payload | Optional structured input the run starts with. Choose Content Type to take the fields of an existing content type, or Custom Shape to define the fields here. Leave unset—shown as None—to start with no input. |
| MCP Tool | On the Advanced tab. Offers this trigger to AI agents as a tool on the workspace's MCP server, with the payload as the tool's input; see Offering it to AI agents. Leave empty to keep it off the tool list. |
What later steps see
With a payload set, the trigger's output is that payload's fields, filled in for each run. With none, the trigger produces no output.
A file uploaded on the run form reaches later steps as a file—its storage, path, content type, and public URL—shown in the variable picker with its filename. An uploaded file belongs to the run that uploaded it and is deleted when that run is archived. Starting a new run from a past run's values gives the new run its own copy of each file, so archiving either run leaves the other's intact. A file given by URL instead is used where it is: it belongs to no run, and later steps read whatever is behind the URL when they run.
Running one by hand
The node offers Run Automation and Run as Dry Run. A trigger with no payload fires immediately; one with a payload opens a short form first, so you fill in the fields the run will start with. Run as Dry Run rehearses a published automation once without changing anything about it. A draft or a revision always runs as a rehearsal, so on those the two buttons do the same thing.
When the trigger has a payload, the form opens above a row of Recent runs pills—New plus up to five past runs with distinct payloads, each labeled with its run number and a relative time, for example "Run #12 · 3 minutes ago". Click one to prefill the form with that run's values instead of starting blank.
The run buttons appear only on a trigger that is the start of the flow, and only while the automation is enabled.
As a step in the middle of a flow
An On Demand trigger placed mid-flow does not start anything—it holds the run there and waits for a person. While it waits, the node offers Resume, which collects the payload (or simply confirms, when there is none) and lets the run continue. The waiting step also appears in the approvals inbox, where Respond opens the same form.
Running from a content edit page
When an on-demand automation is registered against a content type, editors can run it straight from that content's edit page instead of opening the editor. See Triggers.
Offering it to AI agents
A trigger that starts the flow can be offered to AI agents as a tool on the workspace's MCP server. On the Advanced tab, add an MCP Tool and give it a Name—a lowercase letter, then lowercase letters, digits and underscores, up to 40 characters, unique in the workspace; left empty, it is filled in with the automation's name in that form when you save—and a Description, which defaults to the automation's. The name is stored, so renaming the automation later does not break the agents already calling the tool. On the root server, which offers every workspace, the name is prefixed with the workspace's code. A call answers as soon as the run has started, with the run's handle; Wait For Completion makes it wait up to 10 seconds for the run's result instead, for an automation that answers quickly.
The tool's inputs are the payload's fields, when the payload is a Custom Shape; each field's Description is what the agent reads about it. A Content Type payload, or a Custom Shape holding a File field, cannot be offered, because an agent has nothing to send for either, and the form says so. The field is hidden on a trigger in the middle of a flow, which is never offered.
The automation has to be on for the tool to be listed. A draft that is on is listed and rehearses, which is how you test a tool before publishing. What the agent gets back, and how it reads a run that is still working, is covered in Results and MCP tools.
Seeing what calls it
Another automation can start this one through an Automation action targeting it. When something does, the trigger's Used By tab lists each calling automation, its state, which of its steps fire this trigger and whether that step waits for the run to finish, and when it last called the trigger, linked to that run, with its call count over the last 30 days. A trigger offered as an MCP tool names the tool there too. A trigger nothing calls says so.
Removing a trigger something still calls, or that is offered as a tool, or switching it to a different trigger type, asks for confirmation first.