Glossary · IT & systems
Tool calling
The technique by which an AI model indicates which tool or function should be run, and with which data. The application around the model carries out that call and returns the result to the model.
Tool calling, also known as function calling, is the way a language model makes something happen outside the conversation. The model is given a list of available tools, each with a name, a short description and the data the tool needs. When a task calls for it, the model does not answer with text but with a structured call, for example to look up the delivery time of a particular item.
The model does not carry out that call itself. The application around the model does that, gets the result back and passes it on to the model. The model then uses it to formulate an answer, or picks the next tool.
Tool calling, MCP and API
Tool calling is the mechanism inside the model. MCPAn open standard that lets AI applications such as Claude, ChatGPT, Gemini or your own AI agent discover for themselves which tools a system offers, and then call those tools. An MCP server usually sits on top of an existing API. is the standard that sets out how tools are offered and found, so that a tool built once works in different AI applications. Behind the tool there is usually just an API that does the actual work. That is how tool calling turns a language model into an AI agentAn AI application that doesn't just answer, but takes steps itself to complete a task. It chooses the tools it needs, calls them, looks at the result and decides what the next step is..
Why it is worth knowing
Whoever decides which tools a model gets decides what it can do. A model with only read tools can look things up and summarise. A model with write tools can place orders or change data. That choice determines what you entrust to an AI. So make it deliberately, together with whoever manages your systems.
