Invoke Agentic Action
The Invoke Agentic Action action runs a configured Agentic Action. It evaluates your prompt, chooses from its configured actions and remote MCP tools, and returns a final response.
When to Use This Action
Use this action when:
- A task requires decisions based on runtime context
- The exact sequence of tool calls cannot be defined in advance
- The selected Agentic Action uses tools from a connected Streamable HTTP MCP server
- You want to reuse an Agentic Action from a Hosted Mailbox, Remote Mailbox, Scheduled Task, Form, Macro, or another Agentic Action
Don't use this action when the required steps and their order are already known. Configure those actions directly or use a Macro for deterministic, reusable automation. For a single AI request that does not need tools, use Prompt AI.
How It Works
- The action sends your prompt to the selected Agentic Action.
- The Agentic Action evaluates the prompt and its configured instructions.
- It calls configured actions or allowed remote MCP tools as needed.
- It returns the final text for Simple responses. For Structured responses with an output schema, it returns an object with the properties that schema defines.
- The action stores the final response in the variable you configure.
History keeps each tool call together
The Invoke Agentic Action step contains the Agentic Action's activity. Each action or MCP tool call appears once beneath it, with that tool's actions and result grouped together. If a tool runs a macro or another Agentic Action, its work stays inside the same tool-call group instead of appearing as a second copy of the call.
Tool entries show the inputs and result for that invocation. Password inputs defined by a macro or nested Agentic Action stay masked. After each tool finishes, the calling automation's data remains unchanged except for the final result stored by Invoke Agentic Action.
Configuration
| Field | Description |
|---|---|
| Agentic Action | Select the Agentic Action to run. Configure its instructions, tools, and response format before using this action. |
| Prompt | Enter the task or question for the Agentic Action. This field supports text expressions such as {{email.subject}}. |
| Expanded Response | Wrap the final response in a Response property. Enable this when later steps need to reference {{custom.<variableName>.Response}}. |
Returned Variables
Configure Store the results in Variable to use the response in later steps.
| Setting | Returned value |
|---|---|
| Expanded Response disabled | The Agentic Action's simple text response or structured object. |
| Expanded Response enabled | An object whose Response property contains the simple text response or structured object. |
For Structured responses, the returned object follows the output schema configured on the selected Agentic Action. If the Agentic Action does not have an output schema, the action returns plain text for compatibility.
MCP server credentials stay encrypted and do not appear in the returned value or action history.
FAQs
Do I need an external AI account or API key?
No external AI account or API key is needed. This action is in limited release; contact your account manager to request access.
Why did the Agentic Action choose a different tool sequence than expected?
Agentic Actions choose tools at runtime and are non-deterministic. Refine the Agentic Action's instructions and tool selection when it needs clearer boundaries. Use a Macro or direct action sequence when every step must run in a fixed order.