Creating Custom Tools for CrewAI

Last Updated : 01 Sep, 2025

CrewAI is a framework that allows multiple AI agents to work together to complete tasks. One useful feature is the ability to create custom tools that agents can use to perform specific functions. Custom tools are needed when agents need to do something that is not built into CrewAI by default such as performing calculations, accessing APIs or processing data.

By creating a tool, we give an agent the ability to handle these tasks automatically, instead of hardcoding everything in the agent’s logic.

1. Setting Up the Environment

Before creating custom tools, we need to set up the environment.

Installing the CrewAI Package

We will use pip to install the CrewAI package:

!pip install crewai

Setting the API Key

We need to set the API key since some tools require external services like OpenAI.

To know how to extract OpenAI API key refer to: How to find and Use API Key of OpenAI.

Python

import os
os.environ["OPENAI_API_KEY"] = "your-api-key-here"

2. Creating a Custom Tool

We will be making a custom tool in CrewAI using the @tool decorator. This decorator marks a function as a tool that can be used by agents. As an example, we will be making a calculator tool:

@tool("calculator"): Registers the function as a CrewAI tool named "calculator".
query: str: Input string that contains the math expression.
eval(query): Computes the mathematical result.

Note: eval is unsafe for untrusted input. In production, a safer parser should be used.

Python

from crewai.tools import tool

@tool("calculator")
def calculator(query: str) -> str:
    """A simple calculator tool that evaluates math expressions."""
    try:
        result = eval(query)
        return f"The result of {query} is {result}"
    except Exception as e:
        return f"Error: {str(e)}"

3. Defining Agents That Use Tools

Agents are the entities that perform tasks. We can give them access to custom tools to expand their functionality.

role: Defines the agent’s function.
goal: Explains what the agent is trying to achieve.
backstory: Provides context or skills of the agent.
tools: List of tools the agent can use.
verbose: If True, the agent provides detailed reasoning during execution.

Python

from crewai import Agent

researcher = Agent(
    role="Researcher",
    goal="Perform research and compute results using tools.",
    backstory="Handles calculations and prepares data for reporting.",
    tools=[calculator],
    verbose=True

writer = Agent(
    role="Writer",
    goal="Summarize the research into a readable article.",
    backstory="Takes calculation results and writes summaries for users.",
    verbose=True
)

4. Assigning Tasks

Tasks tell agents what to do. Each task is linked to an agent and specifies the expected output.

description: Explains the task clearly.
agent: Assigns which agent will perform the task.
expected_output: Specifies what the agent should return.
context: Allows the agent to access previous task outputs.

Python

from crewai import Task

task1 = Task(
    description="Calculate '12 * (5 + 3)' using the calculator tool.",
    agent=researcher,
    expected_output="The numeric result of the expression."
)

task2 = Task(
    description="Summarize the researcher's calculation in a short article.",
    agent=writer,
    expected_output="A summary article including the calculation result.",
    context=[task1]  # Provides the output of task1 as context
)

5. Creating and Running a Crew

A Crew groups agents and tasks so they can work together on a goal.

agents: List of agents participating in the Crew.
tasks: List of tasks assigned to the Crew.
process: Determines execution order (sequential or parallel).
kickoff(): Starts the Crew and executes the tasks.

Python

from crewai import Crew, Process

crew = Crew(
    agents=[researcher, writer],
    tasks=[task1, task2],
    process=Process.sequential,  # Executes tasks in order
    verbose=True
)

result = crew.kickoff()
print("\nFinal Result:\n", result)