How a model acts beyond text: it emits a structured call to an external tool such as search, code, or an API; a harness runs it and feeds the result back. The mechanism under AI agents.
Tool Use, also called function calling, is how a language model acts beyond generating text: it emits a structured request to call an external tool, a program runs that call, and the result is fed back into the model's context to continue from. Search, a calculator, code execution, a database query, or a booking API can all be tools.
The model does not run anything itself. It is given a list of available tools with their inputs, and when a task needs one it outputs a call, the tool name and arguments, in a fixed format; the surrounding agentic harness executes it and returns the output, which the model reads like any other part of its prompt. This loop of call, execute, read, continue is what turns a text predictor into an AI agent.
Tool use is the mechanism under most of the agent stack. A live web search to ground an answer is a tool call; the Model Context Protocol is a standard way to expose tools to a model; and WebMCP brings the same idea to websites. For AI visibility this is where the agentic web is headed: brands reached not through a page a person reads but through a tool an agent calls, a different surface to be present on.
