QAI tutorial

What you can do here

• Do you ask questions to artificial intelligence (AI) assistants?
• Do some of your questions ask for technical information?

If you answered both of these questions with yes, this QAI tutorial is designed for you.

AI needs your help

When you get AI help, you use AI platforms to interact with large language models (LLMs). LLMs are superb at understanding your questions, fetching information for you, and formulating answers from that information. But LLMs are not good at performing technical tasks to produce the accurate information they need. For those tasks, LLMs use tools. If you ask an LLM a question that requires a tool, you cannot assume that the LLM will know how to discover that tool. Typically, the LLM will try to answer you without the tool and will give you an inferior answer: superficial, vague, incomplete, inaccurate, or even fabricated. To ensure that the LLM uses the right tool, you can tell the AI platform to make the tool available to its LLMs.

This tutorial shows you how to do that.

Let’s begin with a concrete example

Suppose the AI platform you want to use is Claude, and you will use it in a web browser.

Also suppose you will be asking an LLM on that platform to give you technical details about software quality: specifically, about the quality of a particular web page.

For that task, the LLM will need a tool that can test a web page to discover quality defects and report those defects to the LLM.

Finally, suppose you know about a tool that has that capability: Kilotest. You also know the URL that Kilotest makes available to LLMs: https://kilotest.com/mcp.

What to do

You can solve this problem with two steps.

Step 1: Become a user

If you are not already a user of the Claude AI platform, register yourself on it as a user. You can do this for free. Navigate to https://claude.ai/login and use either Google or your email address to create an account. After you proceed through the substeps of creating a Free plan account, you are on the home page of a new task.

Step 2: Add a connector

Now you are ready to add a connector to the Claude AI platform.

The page has a toolbar on the left side. Click it to expand it, so you can see labels, not only icons.

Now click the Customize tool.

A dialog opens. In the Customize list, click Connectors.

The dialog changes to a Connectors view.

In the Add dropdown list, click Add custom connector.

A subdialog opens, titled Add custom connector. It contains 4 inputs.

• Enter Kilotest into the Name input.
• Enter https://kilotest.com/mcp into the Remote MCP server URL input.
• Leave the OAuth inputs empty.
• Submit the form by clicking the Add button.

The dialog changes to a tool permissions view for the Kilotest connector. You can decide whether an LLM needs your approval to use each of the tools that Kilotest makes available.

Close the dialog by clicking the x button in the top-right corner or by pressing the Escape key.

Finished!

Now the platform will make the connector available to its LLMs.

Here is a simple test for you to verify that your configuration works:

• In the How can I help you today? input, choose which model you want to use.
• Enter the question: What quality issues exist on the World Wide Web Consortium home page?
• Click the Send message up-arrow button or press the Enter key to submit your question.
• From the answer that you get, determine whether the model used any Kilotest tools for this task.

Did this work?

If the model used any Kilotest tools, congratulations! You know how to add tool power to the LLMs on one of the major AI platforms.

But what if it did not work? If an LLM fails to use the Kilotest connector when it should, or if you need to explicitly tell the LLM to use the connector, something is wrong, and the QAI project wants to know about your experience. Please use the comment form to tell us about it.

You can also submit an issue on the QAI GitHub repository.