Tool

The word 'tool' in AI has a specific technical meaning layered over its ordinary meaning. In the LLM sense, a tool is a named function with a defined schema (description, inputs, expected outputs) that the model knows about and can choose to invoke. When a model calls a tool, the calling code (the scaffold) executes the function and returns the result. Common tools: web search, code execution, database query, file read/write, API call.

Tools are what separate a pure text generator from a useful agent. Without tools, the model can only work with what's in its training data and current context. With tools, it can look things up, verify facts, take actions in external systems, and adapt to real-world state it couldn't know from training alone. This is why tool use is central to the agentic paradigm.

The MCP (Model Context Protocol) is a standard for defining and exposing tools to AI models, making it easier for models from different providers to use the same tools. When you see a model described as having 'tool use' or 'function calling' capabilities, that means you can give it a list of tools and it will decide which to call and when, without you hardwiring that logic yourself.