# What is tool use in AI? Giving a model a list of things it can ask to have done, such as search the web or read a file. The model asks; the program around it does the work and hands back the result. - Page: https://tecyfy.com/concepts/tool-use - Subject: AI agents (https://tecyfy.com/concepts) - Level: beginner - Also called: Tools - The lesson: Inside an AI agent: what happens after you press Enter (https://tecyfy.com/learn/ai-agents/inside-an-ai-agent) - Last checked: 2026-10-14 On its own, a language model can only write text. It cannot look up today's weather, open your spreadsheet or send an email. Tool use is how it gets past that. The model is given a list of tools, and when one would help, it asks for it. ## How a tool call works Each tool is described to the model like a card: a name, a sentence on what it does, and the inputs it takes. A weather tool might be "get_weather: the current weather for a city. Input: city name." You ask "Do I need an umbrella in Leeds?". The model sees the card and, instead of guessing, replies with a short structured request: get_weather, city Leeds. Then it stops. Here is the step people miss. **The model does not run the tool.** The program around it, the app, your code or the provider's servers, reads that request, calls the real weather service and pastes the result back into the conversation. The model reads the result and writes its answer: "Yes, rain from 3 pm." ## Tools you already use When ChatGPT or Claude searches the web, that is a tool. So is reading an uploaded file, running code to make a chart, or a connector to your calendar. Many of these now plug in through MCP, an open standard Anthropic launched in 2024 so that one tool can serve many AI apps. An [Agent](https://tecyfy.com/concepts/agent) is tool use in a loop: ask for a tool, read the result, decide the next step, repeat. ## Why it matters Tools fix 2 of a model's biggest weaknesses. Search gets it past its [knowledge cutoff](https://tecyfy.com/concepts/knowledge-cutoff). A calculator or code runner gets it past guessing at arithmetic. They also add risk, because now the model's words turn into actions. A web page or email it reads through a tool can contain hidden instructions, which is [prompt injection](https://tecyfy.com/concepts/prompt-injection). So a careful setup gives a model only the tools the job needs, and asks you before any tool that sends, pays or deletes. When an app asks for permission to connect to your mail or drive, it is asking to give its model a new tool. Grant what the task needs and no more. ## Related - [Function Calling](https://tecyfy.com/concepts/function-calling): The developer's name for tool use. The model replies with the name of a function and its arguments in a fixed format, and your code runs it. - [Agent](https://tecyfy.com/concepts/agent): A model that uses tools in a loop, deciding each next step from what the last one returned, until the task is done. - [Function Schema](https://tecyfy.com/concepts#function-schema): JSON schema that describes tool arguments for function calling. - [Prompt Injection](https://tecyfy.com/concepts/prompt-injection): An attack that hides instructions in text a model reads, such as a web page, an email or a file, so the model follows the attacker instead of you. ## Sources - [How tool use works](https://platform.claude.com/docs/en/agents-and-tools/tool-use/how-tool-use-works), Anthropic, Claude API documentation, 2026-09-25. The model never executes anything on its own. It emits a structured request, your code or the provider's servers run it, and the result comes back into the conversation. - [Define tools](https://platform.claude.com/docs/en/agents-and-tools/tool-use/define-tools), Anthropic, Claude API documentation, 2026-09-25. A tool is given to the model as a name, a description and the inputs it takes. - [Web search tool](https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool), Anthropic, Claude API documentation, 2026-09-25. With web search, the service runs the searches and gives the results to the model, which answers beyond its knowledge cutoff and cites its sources. - [Introducing the Model Context Protocol](https://www.anthropic.com/news/model-context-protocol), Anthropic, 2024-11-25. MCP is an open standard for connecting AI apps to data and tools. Anthropic launched it in November 2024. - [MCP architecture overview](https://modelcontextprotocol.io/docs/learn/architecture), Model Context Protocol documentation, 2026-09-25. An MCP server offers tools, data and prompt templates, and one server can serve many apps. - [Safety in building agents](https://developers.openai.com/api/docs/guides/agent-builder-safety), OpenAI API documentation, 2026-09-25. A person should approve consequential actions. - [Mitigate jailbreaks and prompt injections](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/mitigate-jailbreaks), Anthropic, Claude API documentation, 2026-10-14. Instructions can be embedded in third-party content such as a fetched web page, an email or the result of a tool call. Apply the principle of least privilege so a successful injection can do minimal damage.