An agent never looks inside a tool. It decides when to use a tool, and what to send it, from four things: the tool's name, its description, its parameters, and their descriptions. When these are right, the right tool runs at the right moment with the right values. When they are vague, the tool sits unused, no matter how well built it is.
You edit all of them in the builder under Settings: the name, the description, and the Input parameters section.
What the agent reads
-
Name. A short label for what the tool does, like
order-status-lookup. Letters, numbers, dashes, and underscores. -
Description. The agent compares the customer's request against this text to decide whether to use the tool. Say what the tool does and when to use it: "Looks up the current status of an order. Use when a customer asks where their order is."
-
Parameter names and descriptions. The agent uses these to pull the right values out of the conversation. A description should say what the value is and what a valid one looks like: "The order number. Six characters: a country code followed by four digits, like NL4252."
-
Required. The agent will not run the tool without a value for a required parameter. It asks the customer follow-up questions until it has one.
When the agent is not using your tool
Check these first, in this order. They apply even when the tool itself works perfectly, because they decide whether the tool gets picked at all.
-
The description does not say when to use the tool. A description that only says what the tool does gives the agent nothing to match against the customer's request. Add the "use when" part.
-
Two tools have overlapping descriptions. The agent has to choose between them, and it may choose the other one. Make each description state clearly what makes that tool the right choice.
-
The tool is called with wrong values. The agent guessed, because the parameter description did not say what a valid value looks like. Add the format and an example.
-
The tool runs without information it needs. The parameter it needed was not marked required, so the agent skipped it. Mark it required under Settings in the Input parameters section.
Whether the tool itself functions is a separate question. Test that with the debugger; see Testing your tool.