Tool

On its own a model just writes text. A tool is what gives it reach. You define a tool by giving it a name, a description, and a schema for its inputs, then hand that definition to the model. The model can now choose to call the tool, and your code runs the real action behind it.

A tool definition

Here is a minimal tool an agent might expose. Note that the definition is just a description and an input schema. It contains no logic; the actual work lives in your code.

const readFileTool = {
  name: 'read_file',
  description: 'Read a UTF-8 text file from the working directory',
  input_schema: {
    type: 'object',
    properties: {
      path: { type: 'string', description: 'Path relative to the repo root' },
    },
    required: ['path'],
  },
}

The description matters more than it looks. The model decides *whether* and *how* to use a tool based on that text, so a vague description leads to misuse and a sharp one leads to the tool being called at the right moments.

Two ways tools get added

• Built in. The agent ships with a core set: read, write, run a command, search.
• Via [MCP](/ai-coding-dictionary/mcp). The Model Context Protocol lets you plug in external tools without modifying the agent, so a database or ticketing system becomes callable the same way a built-in tool is.