We identified the problem in our AI wishlist, and we love to hear more about the specific pain to maximize the alignment with such of users like you. There are many types of dynamic locators in the real world, and we’d like to hear more about your situation. Below are some of the common dynamic manners:
(per your comment) The web app is built with an SPA framework, where the frontend developer focuses on the model view rather than the physical DOM element
Some areas on the web page contain dynamic data (like the news homepage, which lists trending articles)
The attributes of the interaction element (id, name, etc.) include the computed data. For instance, the search input element has name=txt_search_user_role where the _user_role is the dynamic one.
What is your particular pain in identifying the element? Indeed, the Studio has provided advanced features such as Smart Locator and Self-healing to support the user in these common situations (in case you may miss them)
Hi @athifk , thanks for your impactful suggestions. I’d like to understand more about some points to maximize the alignment with the end-user
if the Recording Agent can reliably detect and interact with Blazor components, that would be a game changer since Blazor is not currently on the officially supported framework list.
Could you share the possible tooling capabilities in the .NET Blazor MCP server (for instance, in your testing framework) that allow the Recording Agent to produce highly compatible and readiness scripts aligning with your current project? The purpose of this question is to help us (product development team) explore and employ a suitable LLM model
If the Recording Agent helps us generate reusable test objects, data-driven templates, or shared keywords from natural language descriptions
This might involve the data flow when the working project is transferred to our AI system for various analyses. Do you have any recommendations to remediate the data privacy concern or a suitable agentic working manner?
I can extract text and image content from a given PDF file in my Katalon project . I just would like to know if there would be an AI to assert image/text (in a more flexible way)
As I said, for now, I do it with keywords but I got some issues ( differences in resolution, image indexing, …)
I want to clarify an important distinction: TrueTest does not perform pixel-based or color-based visual regression testing. The approach you’re describing—handling visual noise, pixel shifts, color changes, and ignoring zone - is addressed by our Visual Testing product, which is specifically designed for AI-powered visual validation.
TrueTest takes a fundamentally different approach focused on user behavior analysis rather than visual comparisons:
Element-based Detection: TrueTest captures and analyzes element locators (XPath, CSS selectors, IDs, Smart Locators, etc.) from real user interactions
Dynamic Analysis: Our AI analyzes these captured locators to identify:
Dynamic test objects (elements that change frequently)
Stable test objects (reliable elements for test automation)
Critical user flows that need attention
Behavioral Intelligence: TrueTest focuses on suggesting the most critical flows based on actual end-user behavior patterns on your application, helping you prioritize what truly matters for regression testing
To directly answer your questions:
Visual noise handling (animations, loaders, timestamps): These are handled by Visual Testing, not TrueTest
Ignore regions and visual adaptation: Also features of Visual Testing
TrueTest’s value proposition is in capturing real user journeys and converting them into actionable test automation insights, not visual regression detection
My Thoughts on “Mix manual + prompt: Record part of the flow, then type the rest”
This is an excellent direction for enhancing Katalon Studio’s test creation flexibility. Often, real-world automation scenarios involve test steps that are difficult to capture entirely through recording, especially when handling dynamic data, complex API interactions, or custom UI components.
Allowing testers to record part of the flow (for repetitive UI interactions) and then switch seamlessly to prompt-based or manual scripting for the rest would create the best of both worlds speed and control.
Here’s how this could make testing life easier:
Hybrid Efficiency: Start with recording for repetitive flows like login or navigation, and then prompt Katalon to generate advanced logic or data-driven steps.
Enhanced Readability: Testers could refine recorded steps with natural language prompts to make scripts more readable and reusable.
Reduced Maintenance: If prompts can automatically interpret intent (e.g., “verify the user is redirected to the dashboard after login”), it minimizes dependency on fragile UI elements.
AI-Driven Customization: Combining recording with prompt-based authoring could help testers quickly tailor scripts for domain-specific cases—banking, ERP, or CRM workflows without heavy coding.
In short, a “mix manual + prompt” workflow empowers testers to record fast, customize smartly, and maintain efficiently. It’s a perfect blend of low-code convenience and AI-driven intelligence exactly where modern test automation needs to go.
To make sure we completely understand your context, can you give a code snippet (in Studio or whatever the test framework you’re using) to demonstrate the test flow with multiple browser sessions? And then give us the problem you want to assist with from a particular statement in the code snippet
Could you share the possible tooling capabilities in the .NET Blazor MCP server (for instance, in your testing framework) that allow the Recording Agent to produce highly compatible and readiness scripts aligning with your current project? The purpose of this question is to help us (product development team) explore and employ a suitable LLM model
To give you a bit more context our applications are being migrated from ASP.NET Web Forms to Blazor (Server). From a testing standpoint, there are some challenges that make compatibility particularly important for us:
Dynamic DOM Rendering:
Blazor Server updates the DOM via SignalR connections, and the HTML structure can change asynchronously as components re-render. A compatible Recording Agent would need to detect those dynamic updates and correctly bind test objects even when component IDs are not stable.
Component-Level Targeting:
It would be ideal if the Agent (via the MCP server) could recognize Razor component hierarchies (e.g., , , ) instead of just static HTML elements. Integration with Blazor component metadata or reflection APIs could allow it to produce more stable and meaningful locators.
Selectors & Object Repository Alignment:
Currently, our framework relies on descriptive XPath or Smart Locator with logical naming in Katalon’s Object Repository. The MCP server could expose semantic selectors (e.g. by component name or bound model property) that the Agent could use to generate reusable, maintainable test objects directly.
Data Binding Awareness:
In Blazor, many fields are bound via @bind attributes. If the Recording Agent or MCP server could detect @bind relationships, it could automatically suggest data-driven variables and template parameterization for data-driven testing.
Output Alignment:
The generated scripts should align seamlessly with our existing Katalon project structure and framework design. Test objects and scripts should be placed consistently within our existing folder hierarchy (e.g. within predefined suites or keyword libraries) to maintain structure and version control neatness.
This might involve the data flow when the working project is transferred to our AI system for various analyses. Do you have any recommendations to remediate the data privacy concern or a suitable agentic working manner?
Thank you for asking this question. This is incredibly important for us that our data is kept secure considering the industry that we are in. Given that our project data (test cases, page objects, and environment profiles) may contain sensitive internal URLs or user data, we’d strongly prefer an on-premise or hybrid agentic architecture, such as:
Local Pre-Processing:
The Recording Agent could perform client-side sanitization before sending anything to the LLM — e.g. masking internal URLs, test data, or identifiable strings. But yes I understand this is quite tricky to do and won’t be fool proof is most cases.
Metadata-Only Transmission:
Instead of transmitting full page DOMs or scripts, the Agent could send abstracted metadata (component type, control structure, action intent). This protects proprietary UI and data details while still giving the AI enough context to generate steps.
Private MCP Deployment Option:
For enterprise users, offering an option to deploy the MCP server internally (behind the company firewall) would address most privacy and compliance concerns. The local server could handle sensitive content, while only anonymized prompts or semantic representations are sent to the LLM.
Model Selection Control:
Allowing us to configure which LLM endpoint (e.g. local model, private cloud instance, or vendor-managed LLM) the Recording Agent uses would be ideal. This would let us align the AI workflow with our data governance policies as it evolves or changes with time.
User-Confirmed Transmission:
Finally, a visual confirmation before any project data leaves the local environment would increase trust, something like “Review and approve data to be analysed by Katalon AI.” This would ensure that there is full transparency and user confirmation before data is sent for AI processing.
These are what I could think of. My recommendation would be the Private MCP Deployment Option so that we have full control via the local server and wouldn’t need to second guess the safety of our data by being reliant on an external server.
Hey @athifk, thanks a lot for the thoughtful feedback!
You perfectly captured our goal with the Recording Agent: making automation approachable for manual testers and maintainable for advanced users.
The Recording Agent will come with a completely new look for the Web Recorder, and we can’t wait to share it once it’s ready!
We’re aiming to deliver a better, smoother experience focused on generating clean, user-friendly test scripts, with plans to also support natural language keywords for even faster automation.
Also love your point on reusability, we’re exploring ways to auto-create shared objects, data templates, and keywords from both recordings and prompts.
Your feedback really helps us balance convenience with long-term maintainability.
Great question, and yes, this is something we’re actively tackling.
We’re working on an idea called AI Locator that identifies elements by their semantic meaning, not just attributes like XPath or CSS. So even if the UI changes (React, Angular, etc.), Katalon Studio can still find the right element based on its role and context.
On top of that, the upcoming AI Self-Healing will automatically fix locator issues at runtime, helping reduce flakiness and ongoing maintenance effort.
The Recording Agent will work together with the upcoming AI Locator and AI Self-Healing to handle dynamic elements intelligently.
The AI Locator understands elements by their meaning and context, so it can adapt to UI changes automatically, while AI Self-Healing steps in to fix broken locators during execution. Together, they’ll make tests far more stable and resilient over time.
And yes, MCP will enable context-aware logic to be linked directly to the same test flow.
We’re super excited about this direction and really appreciate your input!
Great points @sbindu.adirajpally! We really appreciate the enterprise perspective; it helps us keep the bigger picture in focus.
We’re designing the Recording Agent with interoperability and scalability in mind. That means:
Tests generated through recording or prompts will be compatible with your existing test repositories.
Regarding CI/CD pipeline integration, it’s definitely worth considering. We’re also exploring batch recording, where a single prompt can generate multiple test scripts at once. Would love to hear your thoughts on this approach.
AI PDF Testing is planned in our upcoming roadmap. We’re working to make it easier for testers to read, compare, and verify PDF files directly within Katalon Studio, no extra tools or complex setup needed.
Great to know. PDF testing is currently cumbersome to setup in katalon as Katalon does not have a inbuilt functionality for PDF testing. This definitely is super exciting.
Thank you all for showing up, asking thoughtful questions, and making our Ask Katalon Anything session such a fun and inspiring space. Your energy truly keeps this community shining!
Big congrats to each of you! You’ve won a special community giveaway gift card
Please DM me (@nghi.phan) here on the forum, I’ll send your gift card straight into your inbox
Thank you again for being such a wonderful part of our community. Every question, every idea, every small moment you share here helps someone else grow, and that’s what makes this space so special
With appreciation, Bella and the Katalon Community Team
Eligible Members
).
Here are our Lucky Draw winners!
You’ve won a special community giveaway gift card 
Please DM me (