Creating Zoom Virtual Agent tools

Tools in Zoom Virtual Agent (ZVA) extend the capabilities of voice agents or chat agents by allowing them to interact with external systems, retrieve information, and perform actions. They bridge the gap between natural language understanding and real-world functionality. Tools enable agents to:

Retrieve and process data from non-parametric sources.
Execute backend actions to fulfill user requests.
Enhance the agent's functionality beyond language generation.

Common use cases:

Location lookup: Use IP-based data to show nearby branches to users.
Information retrieval: Fetch order details or account info using API calls or custom scripts based on the user's query.
Information update: Modify records and confirm the update with the user.

Requirements for creating Zoom Virtual Agent tools

Account owner or admin privileges; or relevant role/privilege
Basic, Pro, Business, Education, or Enterprise account
Zoom Virtual Agent and Zoom Contact Center licenses

How to create Zoom Virtual Agent tools
- API Call settings
- Custom script settings
Best practices for creating Zoom Virtual Agent tools

How to create Zoom Virtual Agent tools

Sign in to the Zoom web portal.
In the navigation menu, click AI Studio then Tool Library.
You will see the following tabs:
- All tool types: View all available tools.
- Built-in: Tools provided by Zoom.
- Custom: Tools you create.
In the All tool types or Custom tab, click Create Tool.
Select a tool type: API Call or Custom Script.
Configure the setting for the selected tool type.
Click Create.

API Call settings

The API Call tool enables Zoom Virtual Agent (ZVA) to interact with external systems by calling APIs to retrieve or update data in response to user queries. Admins can define the API call structure, configure input/output parameters, and instruct the LLM when and how to use it.

Name: Enter a display name for your API Call tool.
Description: Provide a concise explanation of when and why the tool should be used. This helps the LLM understand the appropriate context for triggering the API call. For example, "Use this tool when the user asks for the status of their recent orders."
Endpoint URL: Enter the API endpoint. Use placeholders in curly braces for path variables. For example, /api/user/{userId}
Method: Select the HTTP method: GET, POST, PUT, or DELETE.
Authorization (Optional): Add an authorization method (for example, Bearer Token or API Key) if required by the API.
API timeout value: Specify the maximum time (in seconds) to wait for a response before the call times out.
Parameters (Optional): Define input parameters required by the API. These parameters help ZVA construct the API call dynamically based on user input or predefined values. Click Add parameter and provide the following:
- Name: The name of the parameter as expected by the API.
- Description: Describe the purpose of the parameter. Be specific so the LLM can determine how to supply its value.
- Location: Where the parameter will be placed (Path, Query, Header, or Body).
- Type: The data type (for example, string, number, boolean).
- Value source: Where the value comes from (set manually or collected via LLM).
- Set Value / Value: Define how the value is derived (for example, user.phoneNumber, "12345").
Debug: To test your configuration:
- Click the Run button.
- Input Parameters: Shows the values for the parameters you've defined.
- Result: Displays the raw API response so you can verify correctness and troubleshoot if needed.
Response (Optional): Define how to interpret and use data returned by the API. Click Add parameter and complete the following for each response attribute:
- Name: The key from the JSON response.
- Description: Explain the value's significance. This helps the LLM understand how to use it in its reply.
- Type: The data type (for example, string, number, boolean).
- Map to variable (Optional): Map the response value to a ZVA or skill-level variable for later reference.

Custom script settings

The Custom Script tool allows admins to write and execute custom JavaScript logic to process user inputs, transform data, or integrate with systems where APIs may not be available or suitable. This gives agents flexible, tailored behavior beyond standard API calls.

Name: Enter a display name for your custom script tool.
Description: Describe the scenario in which this tool should be used. Be specific to help the LLM determine when to trigger the script. For example, "Use this tool when the user asks for a personalized greeting based on time of day and user location."
Javascript: Write the logic your tool should execute using JavaScript. This script will be run in a secure, isolated environment.
After writing your script, click the Format code button to improve readability and ensure code structure is preserved.
Parameters (Optional): Define input parameters to make the script dynamic. These values can be passed from the user's input, skill context, or other tools. Click Add parameter and provide the following:
- Name: The variable name used inside your script.
- Description: A clear explanation of what this parameter is and how it is used. Helps the LLM determine what to pass in.
- Type: The data type (for example, string, number, boolean).
- Value source: Where the value comes from (set manually or collected via LLM).
- Set Value / Value: Define how the value is derived (for example, user.phoneNumber, "12345").
Response (Optional): Specify what data the script returns and how it should be used. Click Add parameter and provide the following:
- Name: The key name returned from the script (for example, message, score).
- Description: Explain the meaning or use of the value. This helps the LLM incorporate it naturally into its response.
- Type: The format of the returned value (for example, string, number, boolean).
- Map to Variable (Optional): Assign the output to a reusable variable for later use in skills, responses, or reporting.

Best practices for creating Zoom Virtual Agent tools

Use clear, consistent names and descriptions to improve discoverability and guide the LLM.
Map only essential response attributes needed for the agent’s reply.
Use meaningful variable names to simplify referencing in skills and responses.
Test all API calls with the debug tool before publishing.
Write modular, testable scripts and validate input parameters to prevent errors.
Return only the data required to fulfill the user request effectively.