Implementation:CrewAIInc CrewAI Code Interpreter Tool

Knowledge Sources	CrewAI
Domains	Code Execution, Tool Integration, Security
Last Updated	2026-02-11 00:00 GMT

Overview

CodeInterpreterTool is a CrewAI tool that executes Python code in isolated environments, providing three execution modes: Docker containers (safe), restricted sandbox (fallback), and unsafe mode (direct host execution).

Description

The tool enables agents to run dynamically generated Python code with library installation support. It implements a multi-layered security approach:

• Docker isolation (primary) -- Builds and manages Docker containers with volume mounts for code execution. Automatically builds the Docker image if not found, installs required libraries via pip inside the container, executes code, and cleans up containers after each run.
• Restricted sandbox (fallback) -- The SandboxPython class restricts dangerous imports (os, sys, subprocess, shutil, importlib, inspect, tempfile, sysconfig, builtins) and blocks unsafe builtins (exec, eval, open, compile, input, globals, locals, vars, help, dir). It replaces the __import__ function with a restricted version.
• Unsafe mode (explicit opt-in) -- Executes code directly on the host machine using os.system for library installation and exec for code execution. Only for trusted environments.

The tool automatically detects Docker availability and falls back to the restricted sandbox when Docker is not running.

Usage

Use this tool when agents need to execute Python code for data analysis, mathematical computations, file processing, or programmatic problem-solving. Always prefer Docker mode for production deployments. The unsafe mode should only be used in fully trusted environments.

Code Reference

Source Location

• Repository: CrewAI
• File: lib/crewai-tools/src/crewai_tools/tools/code_interpreter_tool/code_interpreter_tool.py
• Lines: 1-392

Signature

class CodeInterpreterTool(BaseTool):
    name: str = "Code Interpreter"
    description: str = "Interprets Python3 code strings with a final print statement."
    args_schema: type[BaseModel] = CodeInterpreterSchema
    default_image_tag: str = "code-interpreter:latest"
    code: str | None = None
    user_dockerfile_path: str | None = None
    user_docker_base_url: str | None = None
    unsafe_mode: bool = False

Import

from crewai_tools.tools.code_interpreter_tool.code_interpreter_tool import CodeInterpreterTool

I/O Contract

Inputs

Name	Type	Required	Description
code	str	Yes	Python3 code to be interpreted; should include a final print statement for output
libraries_used	list[str]	Yes	List of library names to install via pip before execution (e.g., ["numpy", "pandas"])

Constructor Parameters

Name	Type	Required	Description
default_image_tag	str	No	Docker image tag to use (default: "code-interpreter:latest")
code	str	No	Default code to execute if not provided at runtime
user_dockerfile_path	str	No	Path to a custom Dockerfile for building the execution image
user_docker_base_url	str	No	Custom Docker daemon URL
unsafe_mode	bool	No	If True, executes code directly on host without isolation (default: False)

Outputs

Name	Type	Description
return	str	Output from the executed code (stdout from print statements), or an error message string

Usage Examples

Basic Usage

from crewai_tools.tools.code_interpreter_tool.code_interpreter_tool import CodeInterpreterTool

# Default mode: uses Docker if available, falls back to sandbox
tool = CodeInterpreterTool()
result = tool._run(
    code="import numpy as np\nresult = np.mean([1, 2, 3, 4, 5])\nprint(result)",
    libraries_used=["numpy"]
)

# Unsafe mode (only for trusted environments)
tool = CodeInterpreterTool(unsafe_mode=True)
result = tool._run(
    code="result = sum(range(100))\nprint(result)",
    libraries_used=[]
)