Tool Calling: How an LLM Actually Does Things

On its own a model only produces text. It cannot check today's weather, query a database, or send an email, because none of that is text it can write from memory. Tool calling is how it reaches past that limit.

The trick

Tool calling gives the model a menu of actions it can request. Instead of answering, it can say "run the get_weather tool with city set to Tampa." Your code runs that tool, gets the result, and hands it back. The model reads the result and continues. The model never runs anything itself; it asks, your code acts, the model reads the answer.

The loop

1. You give the model a question and a list of tools.
2. It either answers, or asks to call a tool.
3. Your code runs the tool and returns the result.
4. The model uses that result to answer, or asks for another tool.

That back-and-forth is exactly what turns a chatbot into an agent.

Why it matters

Tool calling is the bridge between a thing that talks and a thing that does. Every AI that books a meeting, runs a query, or deploys a server is doing it through this request-and-response. It is also what MCP standardizes, so any model can use any tool.