AI agents in action: autonomous test generation and MCP with QF-Test 11

In this webinar, we’ll show you how QF-Test 11 integrates AI directly into your everyday testing workflow. From AI instructions in individual test steps to automatically generating comprehensive test cases.

We’ll also demonstrate how to connect QF-Test to external AI agents like Claude Code or GitHub Copilot via the new MCP server, making any application ready for agentic workflows.

Materialien

• Presentation (.pdf)
• Demo suites (.zip)

Summary

AI in QF-Test

QF-Test supports test data generation, test suite generation, and AI instructions, among other things. The approach follows the Bring Your Own Model principle: companies can choose which AI model they want to use. This allows them, for example, to retain control over their data by using internal or self-hosted models. Costs can also be adjusted by choosing the appropriate model.

Connecting AI models

The talk begins by setting up AI support in QF-Test. CLI tools can also be integrated, with many configuration values being determined automatically.

AI test data generation

AI data generation provides a first simple way to verify that the AI configuration is working. QF-Test generates test data, such as a table, based on both a user prompt and a predefined system prompt.

The prompts used for generation can be inspected and customized. For example, the user can specify that generated data should be French wherever possible.

AI suite generation

AI suite generation creates a test design from a natural-language description. The user’s prompt is combined with a predefined system prompt.

As an example, the AI is asked to generate tests for a Swing application:

Generate tests to test the functionality of the 5% discount button. Stay in the “Vehicles” tab and don’t open any dialogs via the menu.

QF-Test generates, among other things, a test description and a test plan. Wherever possible, generated steps should be converted into regular QF-Test nodes, with AI optionally serving as a fallback.

AI instructions

AI instructions allow QF-Test to perform tasks during test execution using natural language, for example:

Open the application’s Info dialog.

These instructions make use of internal MCP tools. The execution and the tools used can be inspected in the QF-Test log.

QF-Test as a tool for AI: MCP

The second major topic is the reverse direction: not only can QF-Test use AI, AI applications can also access QF-Test functionality.

QF-Test exposes capabilities such as component recognition and UI control through MCP (Model Context Protocol). QF-Test effectively provides a toolkit for defining custom MCP tools for virtually any application and making those tools available to an AI.

Example: Chess

A simple chess application demonstrates how QF-Test and AI can be combined. The chessboard itself has no game logic, but QF-Test can move the pieces.

First, the AI is allowed to play several moves. The current position can then be sent to an external API, which calculates the next move. QF-Test receives the result and executes the move. While this works, directly controlling the game through AI can be relatively token-intensive.

The next step is to expose existing QF-Test procedures as MCP tools. Functions for moving a piece, calculating the next move, starting the chess application, and resetting the board are turned into explicitly described tools.

This allows an external AI to call these functions directly and decide, for example, when to calculate a new move or move a piece.

Defining custom MCP tools

Custom tools can be defined directly in QF-Test. Procedures are annotated with MCP-specific metadata and documented with descriptions, parameters, and return values.

The resulting MCP suite can then be integrated into QF-Test. An AI application such as Claude can access the tools through the MCP server and use them during a conversation.

Example: Windows Calculator

Finally, the concept is applied to a different application. A normal QF-Test sequence is first recorded for the Windows Calculator:

1. Enter a number
2. Press plus
3. Enter a second number
4. Press equals
5. Read the result

This functionality is then exposed as an MCP tool. An AI can use the QF-Test calculator tool, for example, to add two single-digit numbers and return the result.

An intentionally incorrect expected result can then be used to demonstrate how QF-Test detects and logs the resulting error.

The MCP server can be operated not only interactively, but also in batch mode.

Conclusion

QF-Test 11 connects AI and automated testing in both directions: AI can assist within QF-Test with generating test data, test suites, and test steps. At the same time, QF-Test can expose its capabilities through MCP as a set of tools for external AI applications.

This makes it possible to use existing QF-Test functionality to enable AI-driven interaction with virtually any application.