What is Tool use?
Tool use is the LLM capability where, instead of producing a normal text response, the model emits a structured request to invoke an external function or API. The host application executes the function, returns the result to the model, and the model continues the conversation using that result.
Also known as: function calling, tool calling, LLM tools
The mechanism in one sentence
You give the model a list of available tools (each described by name, purpose, and JSON Schema for arguments); when the model decides a tool is needed, it returns a tool-call object instead of a text answer; you execute the tool; you feed the result back into the conversation; the model continues.
Why it's foundational
Tool use is what makes agents possible. Without it, LLMs are stuck in their training data; with it, they can search the web, query a database, execute code, send an email, or invoke any function you expose. Every AI agent, autonomous coder, or research bot is fundamentally a loop around tool use.
Tool-use schemas
OpenAI's tools/function_calling schema became the dominant standard and is now copied by most providers. Anthropic's Messages API uses a similar but distinct shape. The Model Context Protocol (MCP), introduced by Anthropic in late 2024 and widely adopted by 2026, standardizes how external services expose tools to LLMs — letting a single tool definition work across providers.
Common pitfalls
Tool descriptions matter enormously — vague descriptions cause the model to misuse the tool or hallucinate arguments. Validate tool arguments on your end; never assume the model produced correct JSON. Implement retries with backoff for transient tool failures. Cap the number of tool calls per request to prevent runaway agents.