Tool use means an AI model asking another program to do a job for it, such as searching the web, checking the weather or adding a calendar entry. It is also called function calling. It is how a chatbot goes from only writing words to getting real information and taking real actions.
Think of a manager with an assistant. The manager cannot reach the filing cabinet, so she writes a note: “Fetch the March invoice.” The assistant fetches it and hands it back, and she carries on. The AI is the manager writing the note. The “tool” is the assistant who actually does the fetching.
In a sentence: “Thanks to tool use, the assistant could check my calendar before suggesting a time to meet.”
What the term means
The AI does not run the tool itself. It asks for it to be run. A program tells the model which tools exist and what each one does. When a question needs one, the model replies with a neat, structured request instead of an answer. Anthropic’s documentation says tool use, “also called function calling”, lets Claude call functions that a developer defines or that Anthropic provides.1
A function is just a small piece of program that does one job. OpenAI also uses both names: “Function calling (also known as tool calling)”.2
A simple example
“What is the weather in San Francisco?” A model cannot know today’s weather from its training. So the steps go like this:
- The app tells the model there is a weather tool.
- The model replies, in effect: “Run the weather tool for San Francisco.”
- The app runs it and sends back the result, such as “15 degrees, partly cloudy”.
- The model turns that into a normal answer for you.
That is the worked example in both companies’ guides, and OpenAI lists the same back-and-forth as its five steps.3
Why it matters to you
Tools are what let an AI act, so they are where the risk sits. A tool that only reads the weather is harmless. A tool that sends email, deletes files or pays bills is not. The tools an AI is given matter more than how politely it is asked; permission tiers for agents explains how to sort them.
The AI can fill in the blanks wrongly. If you leave out a detail a tool needs, the model may ask, or it may guess. Anthropic’s own guide gives the example of asking “What’s the weather?” with no place named.1 With a weather tool that is a small slip. With a payment tool it is not, which is why anything that cannot be undone should wait for a person to say yes.
Often confused with
| Term | How it differs |
|---|---|
| AI agent | Differs: tool use is one request for one action. An AI agent chains many tool calls together towards a goal. |
| MCP | Differs: tool use is the AI asking for an action. MCP is a shared standard for plugging whole apps of tools into an assistant. |
| API | Differs: an API is the doorway a program uses to reach a service. A tool often uses an API behind the scenes. |
Questions people ask
Is tool use the same as function calling?
Yes, they are two names for the same idea. Anthropic says “Tool use (also called function calling)”1 and OpenAI says “Function calling (also known as tool calling)”.2
Does the AI run the tool itself?
No. The model asks for the tool; the app runs it and sends the result back. Anthropic’s guide describes the step: “your code runs the lookup, and a second request sends the result back”.3
What is the difference between tool use and MCP?
Tool use is the AI asking for a tool to be run. MCP is a shared standard for plugging tools and data into AI apps in the first place; its makers call it “an open-source standard for connecting AI applications to external systems” and compare it to a USB-C port.4
Learn more on this site
- Agent guardrails — four lines to write before an AI is allowed to act.
- AI agents for beginners — what tool-using AI can do, and how to stay safe.
- Human in the loop meaning — the person who approves before anything risky happens.
- The AI glossary (see “Function calling”) and AI for beginners.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. In tool use, who actually runs the tool?
- The AI model runs it inside itself
- You run it by hand each time
- The app runs it and sends the result back
Show the answer
Answer: The app runs it and sends the result back. The page says the model only asks for a tool; the app runs it and sends the result back to the model.
2. Why does the page say tools are where the risk sits?
- Tools make the AI type more slowly
- Tools let an AI take real actions
- Tools stop the AI from answering
Show the answer
Answer: Tools let an AI take real actions. The page says a tool that reads the weather is harmless, but one that sends email, deletes files or pays bills is not.
3. What should happen before an action that cannot be undone?
- A person should say yes first
- Nothing, if the request was polite
- It should run first and be checked later
Show the answer
Answer: A person should say yes first. The page says the AI may guess a missing detail, so anything that cannot be undone should wait for a person’s yes.
Related terms: Agentic AI · Schema / structured output · Semantic search · Subagent · Vector database · Vibe coding
Related on this site: What Is Function Calling in AI? · What Is Structured Output in AI?
- 1Anthropic, Tool use with Claude, read at source 1 Oct 2026“Tool use (also called function calling) lets Claude call functions that you define or that Anthropic provides. Claude determines when to call a tool based on the user’s request and the tool’s description.” ↑a“Claude (particularly Claude Sonnet) might guess values you didn’t supply”. ↑b“Tool use (also called function calling) lets Claude call functions that you define or that Anthropic provides.” ↑c
- 2OpenAI, Function calling, read at source 1 Oct 2026“Function calling (also known as tool calling) provides a powerful and flexible way for OpenAI models to interface with external systems and access data outside their training data.” ↑a“Function calling (also known as tool calling) provides a powerful and flexible way for OpenAI models to interface with external systems and access data outside their training data.” ↑b
- 3Anthropic, Tool use with Claude, read at source 1 Oct 2026, which walks through a get_weather tool“your code runs the lookup, and a second request sends the result back”. OpenAI, Function calling, read at source 1 Oct 2026, whose flow ends: “Receive a final response from the model (or more tool calls)”. ↑a“your code runs the lookup, and a second request sends the result back”. ↑b
- 4Model Context Protocol, What is the Model Context Protocol (MCP)?, undated documentation, read at source 12 Oct 2026“MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems.” and “Think of MCP like a USB-C port for AI applications.” ↑
The comparisons on this page (the manager and her assistant) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK