Use an agent or workflow as a component-Alibaba Cloud Model Studio(Model Studio)-阿里云帮助中心

You can publish agent or workflow applications as modular components. Other applications can connect to and use these components. This promotes feature reuse and accelerates the development of complex AI applications.

Getting started

This guide uses a weather query component as an example to demonstrate how to develop and use components. You can connect this component to various AI applications, such as travel planning, agriculture management, and outdoor sports, to provide convenient weather query services.

Step 1: Create an application

You can create an agent or workflow application to implement a weather analysis module.

Agent application

Create an application: Go to Application Management and create a new agent application. Configure the agent with the following settings:

Configuration item	Configuration
Model	Qwen-Max-Latest
Prompt	You are a weather analysis assistant. Your task is to query the real-time weather and temperature trends for a specific city.
MCP service	Amap Maps (to get weather information), QuickChart (to create charts)

Note

MCP is a protocol framework that lets you connect external tools to an agent application. The Large Language Model (LLM) can then call these tools. In this example, the tools are Amap Maps and QuickChart. For more information, see MCP introduction.

The MCP services used in this example are free for a limited time.

Test the result: In the dialog box, enter "Query the temperature change trend for Hangzhou." The agent outputs a temperature line chart.

Workflow application

Create an application: Go to Application Management and create a new task-based or conversational workflow application.
Configure the start node: Delete the default city and date parameters because they are not used in this example.

Configure the LLM node: Drag an LLM node onto the canvas. Connect the start node to the LLM node. Keep the default settings, except for the following configurations.

Configuration item		Configuration content
Model Configuration		Qwen-Max
Parameter settings	Maximum response length	8192
	Enable search	Enabled
	Search policy	Enhanced
	Force search	Enabled
	Search extension	Enable
System Prompt		You are a weather analysis assistant. Your task is to query the real-time weather and temperature trends for a specific city.
User Prompt		Enter '/' to open the variable menu and select "System Variable/query".

Configure the end node: Connect the LLM node to the end node. Keep the default settings, except for the following configurations.

Configuration item	Configuration
Output mode	Text output
Text box	Enter '/' to open the variable menu and select "LLM_1/result".

Test the result: In the test window or dialog box, enter "Query the temperature change trend for Hangzhou." The workflow outputs a text-based forecast.

Step 2: Publish the application as a component

You can publish components when you publish applications. Additionally, for applications that are already published, you can also publish them as components in the My Components panel.

Publish an application as a component

Publish the application: Go to the editing page of the agent or workflow application, and then click Publish Application.
Publish components: When publishing an application, select Publish Application Component.

You can also create components at any time in the Publish Channel.

Publish a component from the management panel

The console provides Component Management. You can click Create Component to publish existing agents as components. You can also view, modify, or delete application components.

Step 3: Configure the component name and parameters

When you publish a component, you must set a suitable name, description, and parameters to connect it to an agent or workflow application.

To standardize the connection type, the component has the following preset system parameters:

query: This parameter passes the text entered by the user. For example: "Please get me the weather forecast for Hangzhou today."
imageList: This parameter passes a list of public URLs for images entered by the user. This parameter is valid only when the component uses an Image and Video Understanding model.

Note

Preset system parameters cannot be deleted. If you do not need a preset parameter, hide it when you configure parameters (set Visibility to No).

Enter the component name and description

Component name:

A name that describes the component's function. (Example: Weather Analysis Agent)
Component description:

A brief introduction to the component's function. For example: Used to analyze weather information for a city.
- When connected to an agent, the LLM automatically determines whether to call the component based on its description and context.
- When connected to a workflow, this description only explains the component's function and does not affect the workflow's operation.

Set input parameters

Configuration item	Purpose		Example
Input information	Parameter name	The original name of the parameter. It cannot be modified.	query	imageList
	Alias	To avoid duplicate parameter names, you must create an alias for the parameter. The caller of the component can only see the alias.	userQuery	imageList
	Parameter description	A description of the parameter's content. When a component integrates with an agent and the method is Model Recognition, the Large Language Model (LLM) automatically populates the parameter values based on the Description. In other cases, this description is for informational purposes only.	A query instruction in natural language	Image links uploaded by the user
	Visible	Specifies whether the component exposes this parameter. Use this to hide unneeded parameters. For example, if the component uses a text model, hide the image input parameter.	Yes	No
	Parameter passing method	Business Pass-through: When connected to an agent, the value is provided by the user of the component. When connected to a workflow, the value is provided by an ancestor node. Model Recognition: When connected to an agent, the LLM automatically infers and fills the value from the context based on the parameter description. When you integrate with a workflow, even if the parameter's Passing Method is set to Model Recognition, the application will not automatically infer the parameter value. You must explicitly provide an input value for this parameter from an ancestor node, just as you would for Business Pass-through.	Model recognition	Model recognition

Step 4: Connect the component to an agent or workflow application

Agent application

You can connect a component to an agent as a tool. In this scenario, the agent automatically uses the component based on the user's query.

If the integrated component contains parameters that require Business Pass-through, you can manually enter them in Input Variable Configuration during testing, or pass them through the biz_param parameter during an API call.

Connect the component: Go to Application Management and create a new agent application. Configure the agent with the following settings:

Configuration item	Configuration
Model	Qwen-Max-Latest
Prompt	You are an agent. Continue processing until the user's query is fully resolved, then end your turn and return control to the user. Only end your turn when you are certain the problem is solved. If you are unsure about the content of files or the structure of code repositories related to the user's request, use your tools to read files and gather information. Do not guess or invent answers. Before each function call, you must plan thoroughly and reflect deeply on the results of previous function calls. Do not complete the entire process just by calling functions, as this can weaken your ability to solve problems and think insightfully.
Agent	Weather Analysis Agent

Test the result: In the dialog box, enter "Get me the weather forecast for Hangzhou." The agent calls the component and outputs a weather forecast, indicating that the component is successfully connected to the agent.

Workflow application

You can connect a component to a workflow as a component node. In this scenario, you must manually pass the required parameters to the component and its output to the next node.

Create an application: Go to Application Management and create a new task-based or conversational workflow application.
Configure the start node: Delete the default city and date parameters because they are not used in this example.

Configure the component node: Drag a component node onto the canvas. Select the agent or workflow component you created earlier. Connect the start node to the component node. Keep the default settings, except for the following configurations.

Configuration item	Purpose	Configuration
Input	Pass the content that the component needs to process.	System Variable/query
Streaming output	If enabled, the content generated by the component is returned as a stream.	Enabled

Configure the end node: Connect the component node to the end node. Keep the default settings, except for the following configurations.

Configuration item	Configuration
Output mode	Text output
Text box	Enter '/' to open the variable menu and select "Component_1/result".

Test the result: In the test window or dialog box, enter "Query the temperature change trend for Hangzhou." The workflow outputs a weather forecast, indicating that the component is successfully connected to the workflow.

Notes

Automatic component updates: After an application is published as a component, the component is automatically updated each time the application is republished.
Avoid nested calls: If application A calls application B, and application B in turn calls application A, the applications enter a recursive loop, which makes the feature unavailable.
Avoid multi-level calls: Applications have a maximum runtime limit. Therefore, multi-level calls, such as application A calling B and B calling C, can easily cause timeout errors.

.unionContainer .markdown-body table tr { border-bottom: 1px solid #E9E9E9; }

.unionContainer .markdown-body table.table-no-border tr { border: none; }

.unionContainer .markdown-body table tr:last-child { border: none; }

.unionContainer .markdown-body h2 span.ph.tips { color: #666; font-size: 16px;

}