Tool use (function calling)
Tool use, also called function calling, is the mechanism by which a language model invokes external functions: the host application declares tools with typed schemas, the model emits a structured call with arguments, and the host executes it and returns the result into the model's context. It is how agents act — calling APIs, querying databases, running code — rather than only generating text.
Tool declarations have to come from somewhere, and for APIs they come from the machine-readable surface: OpenAPI operations map naturally onto tool definitions, and MCP servers declare their tools with JSON Schema-typed inputs. An API with a clean, parseable spec converts directly into tools an agent can call; an API documented only in prose forces the agent to reconstruct the schema by reading, with every misreading becoming a malformed call.
Correct tool use is downstream of comprehension. A model that misunderstands the documentation builds requests with wrong parameters, wrong endpoints, or capabilities the API never had — the hallucinated-integration failure that surfaces later as a support ticket blaming the API. The quality of what the agent read determines the quality of what it calls.
Discry's comprehension dimension measures the precondition for correct tool use: request-construction tasks quiz models on building correct requests from the API's fetched docs, graded mechanically against citation-verified ground truth, while trap tasks probe whether a model invents capabilities the API does not support. Each profile reports its request-task coverage.
What does an AI agent make of your API?
Find out in about a minute — no signup.
Discry your API — free