Tool calling is the capability that allows AI models to interact with external tools, APIs, databases, and systems to perform actions beyond text generation. The process follows a structured loop: the model recognizes when a request needs an external tool, selects the appropriate one from a schema-described set, generates a structured JSON request, and incorporates the tool's response into its final answer. Different tool types include data retrieval (including RAG), code execution, action tools (email, CRM updates), and IoT/physical system integrations. Tool calling is distinguished from function calling — the latter being a subset introduced by OpenAI in 2023, while tool calling is the broader industry-standard term. Databricks supports tool-calling agents through Agent Bricks, which natively integrates MCP, Unity Catalog governance, and popular frameworks like LangGraph and CrewAI. Gartner predicts 40% of enterprise apps will embed task-specific AI agents by end of 2026, up from under 5% in 2025.
Table of contents
Why tool calling matters for AI agentsHow does AI tool calling work?Common types of tool callingQuestions this post answers
What is the difference between tool calling and function calling in LLMs?
Function calling is a subset of tool calling. Function calling refers specifically to a model generating structured arguments for a predefined function — a term introduced by OpenAI in 2023. Tool calling is the broader concept encompassing API integrations, database queries, code execution, and any external interaction. The industry has converged on 'tool calling' as agents have grown to orchestrate more complex multi-step workflows. Teams deciding which term to use in their agent architecture find the distinction matters when reading across different platform docs on daily.dev.
How does the tool calling loop work in an AI agent step by step?
An LLM with tool-calling support follows four steps: it recognizes the user's request requires an external tool, selects the appropriate tool from a schema-described set, generates a structured JSON request with the correct parameters, and incorporates the tool's response into its final natural-language answer. The model never executes tools directly — an application layer handles execution and passes results back as context. Developers building their first agent workflow track patterns like this across frameworks on daily.dev.
What is the Model Context Protocol (MCP) and how does it relate to tool calling?
MCP (Model Context Protocol) is a standardization layer for how tools are described, discovered, and invoked by AI agents. It provides a consistent interface so agents can connect to any MCP-compatible tool without writing custom integration code for each one. Tool calling is the capability; MCP is the open standard that makes tool integrations portable and interoperable across different agent frameworks and platforms. Engineers evaluating MCP adoption for their agent stack follow ecosystem developments on daily.dev.