06 · Agents, Tools & MCP · Topic 03

What is Tool Calling?

Tool calling lets an AI application ask an external function or service to perform a defined action.

What is it?

Tool calling is a structured way for a model to request an action from an external function or service.

The application defines each tool, its inputs, and its permissions. The model proposes a call, but the application validates and executes it.

Think of it like...

Think of it like filling out a service request.

The request names the service and required fields. A separate system checks the request and performs the action. The model proposes the request; it does not automatically own the external system.

How it works

1 · DefineDescribe the toolDeclare its name, inputs, and limits.

2 · RequestChoose a callThe model returns a structured tool request.

3 · ValidateCheck permissionsThe application checks inputs and authorization.

4 · ExecuteRun the toolThe external function or service performs the action.

5 · ReturnSend the resultTool output returns to the application or model.

Real-world examples

Everyday example · Calendar event

Input: “Schedule a meeting tomorrow at 10.”

AI / System: creates a structured calendar request

Output: a confirmed event or an error

Business example · Account lookup

Input: a support request with an account ID

AI / System: calls an approved account service

Output: permitted account data for the workflow

What it is NOT

Tool Calling ≠ Text Generation

Tool calling: requests an external action in a defined shape.

Text generation: returns language as model output.

Tool Calling ≠ Automatic Permission

Tool calling: works within application permissions.

Automatic permission: a model does not receive by default.

Tool Calling ≠ Tool Result

Tool calling: is the request to use a tool.

Tool result: is the output returned after execution.

Remember this

Tool calling connects model decisions to controlled external actions.

Independent explainer

Video

visual explainer