Agents
Tool Use
A large language model's ability to call external tools, functions, and APIs to act beyond text generation; the core capability underlying AI agents.
Tool use is a large language model’s ability to call external functions, tools, and APIs instead of only generating text. Through tool use a model can search the web, execute code, read files, or drive other systems, letting it act in the real world rather than just describe it.
The mechanism runs as a loop: the application supplies tool definitions (name, description, and a parameter schema), the model decides which tool to call and returns a structured request, your code executes the operation, and the result is fed back to the model, which continues reasoning. It is often called function calling; protocols such as MCP standardize how tools are exposed to models.
Tool use is the foundation of every AI agent: without it, a model is limited to its training-time knowledge. By 2025-2026 it became a standard capability of frontier models and is measured on benchmarks like SWE-bench, where access to tools produces outsized gains in capability.