# Invoke an Agent

> Send a message to a remote Agent2Agent (A2A) agent from a Maestro agentic process and use the agent's response in later process steps.

## Description

This activity sends a message to a remote A2A agent and returns the agent's response, so the agent can take part in a process orchestrated by Maestro.

The agent answers in one of two ways:

- **Immediate reply**: the agent returns a message, and the task completes with that message.
- **Long-running task**: the agent returns a task. Integration Service polls the agent for the task status, and the task completes when the agent task reaches a final state. For details, refer to [Agent 2 Agent events](https://docs.uipath.com/integration-service/automation-cloud/latest/user-guide/uipath-google-agent2agent-events).

## Prerequisites

- An Agent 2 Agent connection to the agent you want to call. For details, refer to [Agent 2 Agent authentication](https://docs.uipath.com/integration-service/automation-cloud/latest/user-guide/uipath-google-agent2agent-authentication).
- The agent meets the [remote agent requirements](https://docs.uipath.com/integration-service/automation-cloud/latest/user-guide/uipath-google-agent2agent#remote-agent-requirements).

## Adding the activity to a Maestro process

1. Add a service task element to the canvas and open the task's **Properties** panel.
2. In the **Implementation** section, from the **Action** dropdown list, select **Start and wait for external agent**.
3. Select the **Agent 2 Agent** connector.
4. Select an existing connection or create a new one.
5. From **Activity**, select **Invoke an Agent**.
6. From **Agent Skills**, select the skill to use.
7. In **Message content**, enter the message to send to the agent.

At runtime, Maestro sends the message to the agent and waits for the response before moving to the next step.

## Configuration

| Field | Direction | Description |
| --- | --- | --- |
| **Connection** | Input | The connection established in Integration Service. Access the dropdown menu to choose, add, or manage connections. |
| **Agent Skills** | Input | Required. A skill from the Agent Card of the agent. Selecting a skill shows its description, examples, and parameters, to help you write the message. |
| **Message content** | Input | Required. The text of the message sent to the agent. |
| **Role** | Input | Optional. The role of the message sender, for example `user`. |
| **Context identifier** | Input and output | Optional. The ID of an existing conversation context. Enter the **Context identifier** output of an earlier **Invoke an Agent** task to continue that conversation. |
| **Blocking** | Input | Optional. Whether to ask the agent for a synchronous reply instead of a long-running task. |
| **Accepted output modes** | Input | Optional. The response formats you accept, for example `text`. |
| **History length** | Input | Optional. The number of previous messages to include in the context. |
| **Metadata** | Input and output | Optional. Key-value pairs for additional context or tracking information. |
| **Current status state** | Output | The final state of the agent task, for a long-running task. |
| **Status error** | Output | Error information, if the agent task failed. |
| **Artifacts** | Output | The results produced by the agent task. Each artifact has an identifier, a name, and parts. The **Artifact part text** field holds the text of each part. |
| **History** | Output | The messages exchanged in the task, with their role and text. |

## Using the agent's response in the process

To let the agent's response influence the process, for example to make a decision at a gateway, assign it to a process variable.

1. In Design mode, select the service task on the canvas.
2. Select **Properties**.
3. Under **Output**, select **Add new**.
4. Add a variable of type **String**, for example `agent_response`.
5. For **Value**, select the field that holds the agent's answer from the task's response. For a long-running task, this is usually the **Artifact part text** field.

:::tip
When a later step needs structured data, ask the agent to respond only with a JSON object, then parse the response in Maestro.
:::