Tool calling

Tool calling, also called function calling, is a model capability in which the model returns a structured request to run a named tool with arguments, instead of plain text. The application runs the tool and sends the result back, and the model continues.

What it is

The round trip, as described in Anthropic's tool use docs:

1. The request lists the tools, each with a name, a description and a JSON Schema for its input.
2. The model decides to call one and stops with a tool_use block naming the tool and its input.
3. Your code runs the tool and sends a tool_result back.
4. The model uses the result to answer, or calls another tool.

Some tools run on the provider's side instead, such as web search. Tool definitions and results all count as input tokens.

Why it matters

Tools are how an agent stops just talking and starts acting: reading a file, searching an issue tracker, running a command. Every tool is also a permission, so which tools an agent sees deserves as much care as what it's told.

In AI CodeOps

Agents in a session call tools from the remote MCP servers you add. As the agents see them:

tool names   <server>__<tool>
per session  up to 32 MCP tools
per step     up to 6 rounds of tool calls
per call     30-second timeout; long results are shortened

Tools a policy rule denies, or an agent's permissions exclude, are removed before the model sees the list.