StudioAssist Agent mode
This document introduces StudioAssist’s Agent mode alongside Ask mode, use cases like code generation with MCP servers, and basic setups.
- Katalon Studio Enterprise version 10.3.2 onwards.
- AI model selected for StudioAssist in Preferences > StudioAssist. You can select Katalon's AI provider or your own, with a personal key (see StudioAssist Preferences) or contact your Account Admin (see Configure AI Services).
Think of Ask Mode as “answer my question” and Agent mode as “help me get this done”. Refer to the table below to see how each mode processes your prompt.
| Prompt | Ask mode | Agent mode |
|---|---|---|
| “Create a new test case” | Suggests code | Looks up best practices in Katalon Docs and automatically creates a new test case in the KS app |
| “Explain this test case” | Explains code | Explains code with references to related Katalon keywords and documentation |
| “How to create a test case?” | Gives basic steps | Guides you with the latest step-by-step instructions |
| “Troubleshoot an error at step 3” | Suggests generic fixes | Analyzes the error, checks docs, and proposes validated solutions |
In short, you can achieve more with StudioAssist Agent mode:
- Automate test case creation and editing: Generate new test cases or update existing ones with MCP tools, reducing manual work and saving time.
- Extend automation beyond Katalon: Connect with third-party MCP servers to streamline workflows, pull external data, or integrate with other systems directly inside Studio.
Learn how to connect an external MCP server in Adding an external MCP Server.
Get started - Use Agent Mode
- Open StudioAssist window.
StudioAssist chat window automatically opens on startup, on the right panel:
- Switch from Ask to Agent mode:
- The default mode is Ask when you first enable StudioAssist. As you work, the chat window remembers your last selected mode.
- Chat history is preserved when switching modes. To clear history, start a new chat by turning on/off StudioAssist.
- Start prompting. The more details you provide, the better StudioAssist understands your project context (attaching project files is possible).
You can use the Prompt Library to customize engineering prompts and provide more context for Agent mode. This helps improve the accuracy of StudioAssist responses by tailoring how it generates, explains, or chats about code.
- View results.
If you use Agent mode to create or update a test case, StudioAssist will open the test case once done:
(Optional) Connect to external MCP servers
By default, StudioAssist integrates with 3 built-in MCP tools, allowing basic capabilities:
- Katalon MCP: Connects StudioAssist to Katalon Docs, so your answers are up-to-date and backed by official guides on Katalon Docs.
- Katalon Studio MCP: Connects StudioAssist to your existing project, allowing it to read, create, and edit test cases directly.
- Katalon TestOps MCP: Connects StudioAssist to TestOps, allows direct operations on TestOps objects (Test Cases, Test Suites, Requirements, Sprints, and Releases) from Katalon Studio.
See Katalon's built in MCP servers for more details on these servers and tools.
You can also connect to external mcp servers to allow more capabilities. Learn how to connect an external MCP server in Adding an external MCP Server.
Agent Mode Configurations
Before prompting, you can manage MCP servers to disable tools you don't need. Learn more in Manage MCP servers and tools.
You can also configure StudioAssist behaviors when using MCP tools:
- Tool use behavior
- Max tool calls
If tool use behavior isn't configured, StudioAssist will ask you configure with below options:
| Option | How it works | When to use it |
|---|---|---|
| Allow Once | Allow the tool run one time only. | For one-time requests. |
| Allow this Conversation | Allow the tool run without asking each time until you close or clear the conversation. | For ongoing tasks during your current session. |
| Always Allow | Always allow the tool run without asking, even after you close or clear the conversation. You can reset permissions anytime in Reset tool approval. | For tools you find helpful and use regularly. Avoid using this option for tools that can make that can make sensitive-impact (for example, delete) |
If you select Deny, StudioAssist will stop assessing your prompt:
Attach images and screenshots
Image attachment is available from version 10.4.3 onward.
With image and screenshot attachments, you can provide visual context to StudioAssist prompts for test generation, debugging and failure analysis.
-
Select Agent mode and click the paperclip icon to attach files.
-
Choose Upload from computer and select your image(s).
Image limits and supported formats
- Supported formats: PNG, JPEG, WEBP, GIF
- Max size: 10 MB per image
- Max attachments: 10 files per prompt
-
StudioAssist will return the result according to your request.
The following interactive video shows how to attach images using the file picker.
Additionally, you can paste screenshots or images saved in clipboard directly into the prompt using keyboard shortcuts (Ctrl/Cmd+V).
Common use cases
Refer to the following common use cases and example prompts for guidance.
Create a new test case
Automatically generate a test case and add it to your project.
- Make sure to include your URL, username, and password in the prompt, or define them in the Object Repository using Spy Web or Web Recorder.
- Using an AI-generated test case may result in missing import statements, which can prevent the test case from running. To fix this, press Ctrl + Shift + O (Windows/Linux) or Command + Shift + O (macOS) to add the required import statements automatically.
Prompt Sample
Get relevant test objects from the Object Repository
and create a Katalon Studio test case in this project
to perform the following steps,
using objects that I already captured:
1. Open a browser to the URL [http://demoaut.katalon.com](http://demoaut.katalon.com/)
2. Click the make appointment button
3. Fill in the username 'John Doe'
4. Fill in the password 'ThisIsNotAPassword'
5. Click the login button
6. Close the browser
Update existing test cases
Optimize or restructure existing test cases, for example, perform bulk updates on time-consuming actions such as updating descriptions for test cases or test steps.
Prompt Sample
Get all the test case scripts,
summarize the content,
and then update the description of all test cases in this project.
<Attach or paste your test case>
Learn about a feature by using Katalon Docs tool
Get the latest explanation of a feature.
Prompt Sample
Search the Katalon documentation
and explain what <feature> is and how to use it.
Troubleshoot and fix issues in the test case
Agent mode analyzes the error, checks documentation, and applies fixes.
Prompt Sample
This test case failed, troubleshoot the error
using informationfrom Katalon doc and fix the script.
<paste root cause here>
Create and update feature files for Behavior-Driven Development (BDD)
Automatically create or update BDD Feature files using Gherkin syntax.
Prompt Sample
Using the <upload cucumberinstruction.md requirements file>,
write a complete Gherkin feature file for the login scenario <(User Story 1)
from your Jira ticket>.
Then, create and link the corresponding step definitions (Groovy code)
in the relevant step definition file to complete the BDD test script.
Generate a test case from linked requirements
Automatically generate a test case from a linked requirement (e.g., a Jira ticket) using existing test objects and any additional related information you provide.
Prompt Sample
Retrieve the requirement from <Jira ticket ID> on <Jira instance URL>,
then create a new test case for the <feature name> feature,
ensuring it uses existing test objects from the Object Repository.
Create a test suite of existing test cases
Automatically generate a test suite that groups existing test cases for collective execution, reporting, and maintenance.
Prompt Sample
Create a new regression test that includes test cases I already have.
<attach the test plan or fetch it from Confluence>
Data-driven testing
Goal: Run the same test case multiple times using varied input data.
Step 1: Generate the test data table.
Prompt Sample
Create a test data table to test this login flow,
including username and password for both success and failure cases,
as well as edge cases.
Let me preview the table
before actually creating the test data in KS <attach test case>.
Step 2: Create internal test data
Prompt Sample
Create the test data based on the table above,
only have 2 columns: username and password.
Step 3: Bind the data.
Manually create variables, replace variables in the script, and do data binding, then execute the test (you can use Katalon Docs tool to search the guide if needed).
Create and edit custom keywords
Automatically create or update custom keywords.
Prompt Sample
Add two new keywords in this file <attach file> for 2 actions:
1. Login: include option browser, navigate to URL,
input username and password, then click login.
2. Test datetime picker.
Known limitations
- StudioAssist may not always call tools correctly. When this happens, try again and mention the MCP tool/server, provide more context, or adjust the prompt.
- Changing AI providers requires closing/reopening the chat for the changes to take effect.
- Currently, TestOps MCP Server does not support user login from offline licensing.
- AI providers (OpenAI, Azure OpenAI, Gemini, and AWS Bedrock) currently enforce a limit of
64characters, for the combined length of an MCP Server and its tool name ({server_name}_ _ _{tool_name}). When adding an MCP server, make sure to shorten either the server/tool name (not to be confused with title), or StudioAssist will fail to call these APIs in the case of exceeding.
Troubleshoot connection issues
Sometimes, Agent mode cannot connect to a tool or MCP server, due to network issues for example:
In this case, you can access the server list, read error details by hovering over the warning (!) icon, and choose to refresh the server to try again:
- If you fail to connect to one of 3 built-in Katalon MCP Servers, ensure only one Katalon Studio instance is running.
- AWS Bedrock provider requires that every external MCP tool has a description. Supply tools with description or disable them temporarily, as it can prevent StudioAssist from generating responses.
- Azure OpenAI has a tool limit of 128 per request you can call in a single request. If your AI provider enforces a similar limit, reduce the number of tools or temporarily disable tools you don't need to avoid errors.