Tool use (function calling)
The mechanism by which an AI model calls external functions — declared with typed schemas — instead of only producing text. The difference between an assistant and an agent.
01What it means
Tool use — often called function calling — is the loop by which a model acts on the world rather than only describing it. Your application declares operations with typed schemas: a name, a plaintext description, and the shape of the arguments. The model decides when to call which tool and with what values; your side executes; the result returns into the conversation. A tool is a typed contract between the model and your application — and the model never sees your implementation, only the schema you wrote and the results you returned.
Four situations justify a tool: actions with side effects (sending, writing, updating), fresh or external data the model cannot know (prices, database contents), outputs that must conform to a guaranteed structure, and calls into existing systems. The anti-cases are anything the model can do from its training alone with no side effects — summarisation, translation, general-knowledge Q&A — where a tool round-trip costs more than the work. The borderline heuristic is worth keeping verbatim in spirit: if you are writing a regex to extract a decision from the model's prose, that decision should have been a tool call.
One design fact dominates: the description is the single most important factor in tool performance. The model reads the schema to construct arguments, but it reads the description to decide whether to call the tool at all. Wrong-tool selection is a description problem, not a model-intelligence problem.
02Why it matters for buyers
- —It is the line between an assistant and an agent: chat produces text; tool use produces actions in your systems. What an agent may call is the honest scope of its power.
- —Tool design is an engineering skill, not a model feature: whether the right tool gets called correctly is mostly description-and-schema quality — ask vendors how they test theirs.
- —The tool list is the audit surface: permissions, rate limits, and blast radius live exactly there, which is what a security review should ask to see.
03Common questions
Is tool use the same as function calling?
In practice, yes — "function calling" is the older vendor term for letting a model request a call to a declared function; "tool use" is the agentic framing of the same loop. Expect both words in vendor docs.
Can the model get a tool call wrong?
Yes — arguments can be wrong or missing, which is why inputs are validated server-side and schemas bound the values (enums, required fields). The tool is your code; your rules apply.
What does tool use have to do with MCP?
Tool use is the underlying loop; MCP is the standard way to serve tools to any compliant client. A vendor exposing their capabilities over MCP is offering their tool calls in a form every agent can use.