Tool Use

An LLM's ability to call external functions — search, code execution, APIs — by emitting a structured request the host executes.

term: Tool Use
category: core
short_def: An LLM's ability to call external functions — search, code execution, APIs — by emitting a structured request the host executes.
long_def: The model does not run the tool itself; it outputs a tool call, the host runs it and returns the result, and the model continues. Tool use (also called function calling) is what turns a chat model into an agent.
see_also: ai-agent mcp agentic-loop
etymology_origin: — verify-against-primary-at-build ↗ https://platform.openai.com/docs/guides/function-calling — popularised as 'function calling' by OpenAI (June 2023); no single coiner for the broader 'tool use' sense
related_to: ai-agent mcp agentic-loop
contrast_with: Unlike plain text generation, tool use emits a structured, machine-parseable call that the host actually executes — the model proposes the action; the host performs it.
example: OpenAI shipped function calling for tool use in its API in June 2023; MCP later standardized how those tools are discovered and described across vendors.
source: https://platform.openai.com/docs/guides/function-calling
status: active
why_it_matters: Tool use is the mechanism by which an agent acts on your site; exposing clean, well-described tools (e.g. via MCP) is what makes a site actionable to agents.
sameAs: —
bridge_entity: protocols/capability/mcp
last_verified: 2026-06-15
md_twin: /glossary/tool-use.md