Back to glossary

Term

Tool Call

An invocation of a tool by an AI model during a conversation — such as reading a file, running a bash command, fetching from the web or calling an MCP tool. The foundation of agentic workflows.

Tool Call — explained in more detail

A tool call is the moment when an AI model does not just emit text but actively invokes a tool — reading a file, running a bash command, performing a web search or talking to an MCP server. Based on the context, the model decides which tool it needs and with which arguments.

How it works technically

Instead of plain text, the model emits a structured call with a tool name and parameters. The runtime — Claude Code, an agent framework or a custom client — executes the call and feeds the result back into the context. On that basis the model decides the next step. This is how multi-step, agentic flows emerge instead of single-shot answers.

Safety

Tool calls can be destructive: a bash command deletes files, a SQL tool mutates production data. That is why any agentic environment needs a permission layer — allowlists, a plan mode or an upstream auto-classifier.

Entdecke mehr

Themenuebersicht