For most of AI's recent history, the interaction was one-directional. You gave the model a prompt, it gave you text back. Useful, but limited. The model couldn't check your calendar, query your database, send an email, or do anything that required touching a system outside itself.
That changed when AI models gained the ability to call functions.
Function calling is the mechanism that lets a model recognize when a task requires an external action and then invoke a defined function to perform it. Instead of just describing what a weather lookup would return, the model can actually call a weather API and return real data. Instead of summarizing what a database might contain, it can query the database and work with the actual results. The model still generates the reasoning; the function does the work in the world.
Function calling in LLMs works by giving the model a set of available functions with descriptions of what each one does and what parameters it takes. When the model determines that a function is relevant to the task, it outputs a structured call rather than a text response, the function executes, and the result comes back into the conversation. GPT function calling was an early and influential implementation of this pattern that helped establish how the capability works in practice.
The next step beyond individual function calls is standardized interoperability. Model Context Protocol (MCP) is an open standard that defines how AI models connect to external tools and data sources in a consistent way. Rather than building a custom integration for every tool an AI might need to use, MCP provides a common interface. MCP servers expose tools and resources through that interface, so any MCP-compatible model can use them without bespoke integration work.
This is the foundation of agentic AI: systems that don't just respond to prompts but take sequences of actions to accomplish goals. The articles ahead get into the data connections, access controls, and governance that make this work safely at scale.
