TOOL USE

A model asking its application to run a defined function, such as search or run tests, and using the result.

Zero retention: everything you type stays in this browser. Verify it yourself in DevTools.
Plain-language definition

What Does Tool Use Mean?

Tool use, also called function calling, is how an AI model takes actions beyond writing text. The application tells the model which tools exist, each with a name, a description and the inputs it accepts, such as read_file with a path. When the model decides a tool would help, it replies with a structured request to call it. The application, not the model, runs the tool, then sends the result back for the model to use. Coding agents are built almost entirely from tool use: reading and editing files, running shell commands, searching and fetching documentation.

Why Does Tool Use Matter?

Tool use is what lets models check facts, run tests and change real systems, which makes them far more reliable and far more powerful. It is also where the risk lives, because a tool call can delete files or send data. Clear tool descriptions, sensible permissions and approval for risky calls make the difference between a helpful agent and a hazardous one.

What Does Tool Use Look Like in Practice?

You ask "do the tests pass?" The model replies with a tool call like run_command("npm test"). The agent runs it, gets "2 failed, 48 passed" with the errors, and returns that to the model, which then explains which tests fail and why. Your AGENTS.md told it which command to run. If the command needed approval, the agent would have paused and asked you first.

What Do People Get Wrong About Tool Use?

Models do not run tools themselves or have direct access to your system. They only request tool calls. What actually runs, and with what permissions, is decided by the application around the model.

Which Terms Are Related?

  • Agent loop: The repeated cycle of decide, act with a tool, observe the result, and decide again that makes an AI an agent.
  • MCP (Model Context Protocol): An open standard for connecting AI applications to external tools and data sources through a common interface.
  • Agentic workflow: A setup where an AI takes multiple steps on its own, using tools and checking results, rather than giving one reply.
  • Guardrail: A rule or check, in the instructions or in code, that keeps an AI system from doing what it should not.
  • Prompt injection: An attack where instructions hidden in content an AI reads trick it into acting against its user.

All 28 terms are in the AI glossary. For models side by side, see frontier vs open-weight.

Frequently Asked Questions

Is tool use the same as MCP?

No. Tool use is the general ability of a model to request function calls. MCP is a standard way to package and connect tools so many different AI apps can use the same ones.

Can a model call tools that were not defined?

It can try to, for example by inventing a tool name, but a well-built application only runs tools it defined and rejects anything else. Models are trained to stick to the provided list.

How many tools can an agent have?

There is no fixed rule, but every tool definition takes context tokens, and long tool lists can confuse the model. Many agents load tool details only when they are needed.

Was this helpful?