asset definition

in English: tool definition / tool schema

advancedwidespread

A tool definition is a short, machine-readable description of what a tool is for, when to use it, and what data must be provided for it.

In plain terms

A language model alone only produces text and cannot do anything in the outside world. To write to a calendar, search the web, or query a database, it receives tools. The tool definition is the "user manual" from which the agent learns that the tool exists, what it is good for, and how to call it.

It is like an official form with its header. The header states what matter the form concerns, and the fields indicate what data is requested: name, date, amount, and which are mandatory. The agent fills out this form, while another program handles the actual administration.

A definition typically consists of three parts: the tool's name, a description in everyday language, and the precise structure of the input data, i.e., the schema. The agent decides from the description when to use the tool. From the schema, it knows which fields to fill, and the surrounding program can verify with the same schema whether the filling is formally correct.

It differs from ordinary programming documentation in that the reader is the model itself. If the description is vague, the agent may choose the wrong tool or fill in fields incorrectly, even if the tool itself is flawless.

How it works

The developer writes a definition for each tool: gives it a name, describes its purpose, and lists its parameters with their types (for example, "city: text, required").
The system passes these definitions to the model along with your question in every round, so the agent sees which tools it can choose from. If there are too many tools, the system may load only the descriptions of those currently needed.
When the agent determines that a tool is needed, it does not respond with free text but generates a structured request filled according to the schema: which tool it calls and with what values.
The surrounding program checks whether the request matches the schema, runs the tool, and then returns the result to the model.
The agent continues based on results: it replies to you or calls another tool until the task is completed.

An everyday example

You ask the chat assistant to check if it will rain in Szeged on Saturday. The assistant knows it has a weather-query tool and that it requires a city and a date because this is specified in the tool's definition. If the description omitted the required date format, the agent could easily send an incorrect request.

Who is affected

developerIt writes the definitions, and the agent's reliable use of tools depends on the quality of these definitions. → Code

entrepreneur, executiveTool definitions determine what a corporate agent can access and do, making this a business and security decision as well.

employeeIf the workplace assistant handles something strangely or in the wrong system, the cause is often an inaccurate tool description. → Work & HR

everyoneWhen you attach an extension or connector to a chatbot, you are essentially giving the agent tool definitions.

A common misconception

MythThe tool definition is the tool itself; if the description exists, the agent can already perform the operation.

In realityThe definition is only a description and a form. The actual work is performed by a separate program that someone must write, operate, and authorize.

Limits and risks

The schema only checks the format: a properly filled request can still contain bad data, such as an incorrect date or another client's identifier.
Every definition takes up space in the model's working memory, so too many or overly long tool descriptions make the agent's work slower, costlier and less reliable.
Instructions can also be hidden in the description of a tool from an unknown source, prompting the agent to take unwanted steps, so only trusted tools should be connected.
The definition does not restrict permissions: what the tool can actually delete or send must be configured in the system running behind it.

Similar, but not the same

tool calling – The tool definition is the empty form, while the tool call is the moment when the agent submits it filled out.
API – The API is the actual connection point between programs, while the tool definition is its simplified, model-readable description.
system prompt – The system prompt defines the agent's general behavior and role, while a tool definition describes the use of a specific tool.

Related concepts

Broader

tool use, AI-agent, agent architecture

Narrower

input schema, tool description, parameter

On the edge

tool calling, function calling, Model Context Protocol, JSON schema, structured output

A brief history

The current format spread in 2023 when OpenAI introduced function calling, allowing developers to describe available functions for the model using JSON schemas. Anthropic's Model Context Protocol, published in November 2024, provided a unified, portable format for tool descriptions across providers. By the end of 2025, the protocol came under the stewardship of the Agentic AI Foundation, operating within the Linux Foundation, so it is no longer a standard of a single company.

Where things stand

At the end of 2026, virtually every major model provider and agent framework works with tool definitions consisting of a name, a description and a JSON schema. Attention is focused on writing good descriptions, managing many tools efficiently and filtering out unreliable tools.

Status: 2026 November

Also known as

asset definition · tool description · tool schema · tool definition · tool schema · function definition · function schema · tool specification

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.