# Tools: File Search, MCP and Functions — OpenAI Agent Builder Workflows

Source: https://www.skillbyai.com/en/openai-agent-builder/b-tools

> Give agents capabilities, and describe them well.

## Hosted tools, MCP servers and your own functions

Agents act through **tools**. Typical options include hosted tools such as **file search** over uploaded documents and **web search**, **MCP servers** that expose your systems through the Model Context Protocol, and in code your own **function tools**. The model decides when to call a tool based on its **name, description and parameter schema**, so write them like documentation: what it does, when to use it, what each parameter means. Give each agent only the tools it needs; every extra tool is another way to go wrong.

## A function tool and its generated schema, run

I ran this with Python 3, the openai-agents package 0.22.3 and pydantic 2.13.5, without any API key: it only builds definitions and schemas, no model is called. The decorator turns a typed Python function and its docstring into a tool: the name, the description from the first docstring line, and a strict JSON schema with order_id required and described.

```python
import json
from agents import function_tool

@function_tool
def get_order_status(order_id: str) -> str:
    """Look up the shipping status of an order.

    Args:
        order_id: The order number, for example A1042.
    """
    return "shipped"

print("tool name  :", get_order_status.name)
print("description:", get_order_status.description)
print("schema     :", json.dumps(get_order_status.params_json_schema, indent=1))
```

Output:

```
tool name  : get_order_status
description: Look up the shipping status of an order.
schema     : {
 "properties": {
  "order_id": {
   "description": "The order number, for example A1042.",
   "title": "Order Id",
   "type": "string"
  }
 },
 "required": [
  "order_id"
 ],
 "title": "get_order_status_args",
 "type": "object",
 "additionalProperties": false
}
```

## Return errors as data

Have tools return clear error messages (order not found) instead of crashing, so the agent can recover or ask the user.

**Quiz:** What does the model rely on to decide when to call a tool?

- [ ] The font of the UI
- [ ] The tool's source code
- [ ] The server's IP address
- [x] The tool's name, description and parameter schema

*Answer:* The tool's name, description and parameter schema. Write tool descriptions like good API docs.