What is function calling?
Function calling (also called tool calling) is the mechanism that lets an AI model use external tools: developers first define the available functions by name, description and JSON Schema; when the model decides it needs one, it returns the function to call and its arguments; the application then actually runs it and hands the result back to the model.
The model proposes the call, the application runs it
Take table booking as an example. A developer can define a “check availability” function whose parameters include date, time and party size. The user says “tonight at 7, two people.” The model does not connect to the booking system itself; it returns the function name and its arguments in JSON format. The application receives that, runs the query, and sends the result back to the model, which turns it into an answer. This is an illustrative example.
OpenAI’s documentation calls this flow function calling, also known as tool calling; when a task requires it, the model can propose several calls in succession.
How it relates to MCP and TSO
MCP lets different AI applications list and call tools in the same way, and MCP tools are likewise defined by a name, a description and a JSON Schema for their input parameters.
How clearly a tool’s name, description and parameters are written affects whether the model chooses it and whether it fills in the arguments correctly. That is what TSO checks.
Sources and scope
The Traditional Chinese entry records a source check on 2026-11-04. This English edition does not re-check the sources and adds no new claims.