Skip to Content
User Guide🧩 Available MCP ServersTerminal Sandbox

Terminal Sandbox

Execute code and commands in an isolated sandbox environment. Each chat room has its own independent sandbox, allowing you to safely test code and process data.

Key Features

  • Execute Python, JavaScript, and Go code
  • Run shell commands
  • Create and save files
  • Download result files

Basic Information

Authentication

No authentication required.

Support Scope

Terminal Sandbox is only available in AI Hub chat. It is not supported in 3rd-party MCP clients.

Sandbox Features

Working Directory

  • The home directory is /workspace.
  • Files created in /workspace persist during the session.
  • Files outside /workspace are reset after each execution.

Session Management

  • An independent sandbox is created for each chat room.
  • The sandbox is automatically reset after 10 minutes of inactivity.
  • Files and data are preserved while the session is active.

Supported MCP Tools

execute_code

Execute code in an isolated sandbox environment.

Supported Languages:

  • Python (3.12)
  • JavaScript (Node.js 23.11)
  • Go (1.24)

Input Parameters:

  • code (required): Code to execute
  • language (required): Programming language (python, javascript, go)
  • libraries (optional): List of libraries to install (pip for Python, npm for JavaScript)
  • timeout (optional): Execution timeout in seconds (default: 30)

Return Information:

  • Execution result (stdout)
  • Error messages (stderr)
  • Exit code
  • Execution time

execute_command

Execute shell commands in an isolated sandbox environment.

Input Parameters:

  • command (required): Shell command to execute
  • timeout (optional): Execution timeout in seconds (default: 30)

Return Information:

  • Execution result (stdout)
  • Error messages (stderr)
  • Exit code
  • Execution time

upload_files

Import files uploaded to AI Hub chat into the sandbox’s /workspace/files directory.

Input Parameters:

  • file_ids (required): List of file IDs uploaded to AI Hub chat

export_file

Export files or directories created in the sandbox as downloadable URLs.

Input Parameters:

  • file_path (required): Path to the file or directory to export (e.g., /workspace/report.csv)

Features:

  • Directories are automatically compressed into ZIP files.
  • The returned URL is provided in Markdown link format.
  • URL is valid for 10 minutes.

Usage Examples

Data Analysis with Python

User Input:

Analyze the following product data with pandas and save it as a CSV file:
- Laptop: price $1200, stock 15 units
- Mouse: price $25, stock 50 units
- Keyboard: price $75, stock 30 units
- Monitor: price $350, stock 20 units

Also calculate the average price and total inventory value.

The LLM uses the execute_code tool to generate and execute code similar to the following:

import pandas as pd
 
# Create data
data = {
    'Product': ['Laptop', 'Mouse', 'Keyboard', 'Monitor'],
    'Price': [1200, 25, 75, 350],
    'Stock': [15, 50, 30, 20]
}
 
df = pd.DataFrame(data)
 
# Perform analysis
print(f"Average price: ${df['Price'].mean():.2f}")
print(f"Inventory value: ${(df['Price'] * df['Stock']).sum():,}")
 
# Save to CSV
df.to_csv('/workspace/products.csv', index=False)
print("\nCSV file created.")

Then, the export_file tool provides a download link for the generated CSV file.

JSON Processing with JavaScript

User Input:

Filter users where active is true from the following user data and save as a JSON file:
- Alice (ID: 1, active: true)
- Bob (ID: 2, active: false)  
- Charlie (ID: 3, active: true)

The LLM uses the execute_code tool to generate and execute code similar to the following:

const fs = require('fs');
 
// User data
const users = [
  { id: 1, name: 'Alice', active: true },
  { id: 2, name: 'Bob', active: false },
  { id: 3, name: 'Charlie', active: true }
];
 
// Filter active users only
const activeUsers = users.filter(user => user.active);
 
// Save to file
fs.writeFileSync(
  '/workspace/active_users.json',
  JSON.stringify(activeUsers, null, 2)
);
 
console.log(`${activeUsers.length} active users saved`);

Text File Analysis with Shell Commands

User Input:

Create /workspace/server.log file and add the following content:
2025-12-02 10:15:23 INFO Server started
2025-12-02 10:16:45 ERROR Database connection failed
2025-12-02 10:17:12 INFO Request processed
2025-12-02 10:18:33 ERROR Timeout exception
2025-12-02 10:19:01 INFO User logged in

Then extract only the lines containing ERROR and save to errors.txt.

The LLM uses the execute_command tool to execute commands similar to the following:

# Create log file
cat > /workspace/server.log << 'EOF'
2025-12-02 10:15:23 INFO Server started
2025-12-02 10:16:45 ERROR Database connection failed
2025-12-02 10:17:12 INFO Request processed
2025-12-02 10:18:33 ERROR Timeout exception
2025-12-02 10:19:01 INFO User logged in
EOF
 
# Extract lines containing ERROR
grep "ERROR" /workspace/server.log > /workspace/errors.txt
 
# Check result
cat /workspace/errors.txt

Using Multiple Languages in Combination

User Input:

Calculate project progress rate. 850 out of 1000 tasks are completed.
Save the data as JSON with Python, then calculate completion rate with Go and create a report file.

The LLM executes step by step in a manner similar to the following:

Step 1: Save data with Python

import json
 
data = {'total': 1000, 'completed': 850}
with open('/workspace/data.json', 'w') as f:
    json.dump(data, f)
    
print(f"Data saved: {data}")

Step 2: Generate report with Go

package main
 
import (
    "encoding/json"
    "fmt"
    "os"
)
 
func main() {
    // Read JSON file
    data, _ := os.ReadFile("/workspace/data.json")
    var result map[string]int
    json.Unmarshal(data, &result)
    
    // Calculate completion rate
    percentage := float64(result["completed"]) / float64(result["total"]) * 100
    report := fmt.Sprintf("Project Progress\nTotal: %d tasks\nCompleted: %d tasks\nCompletion Rate: %.1f%%\n", 
        result["total"], result["completed"], percentage)
    
    // Save report file
    os.WriteFile("/workspace/report.txt", []byte(report), 0644)
    fmt.Print(report)
}

Precautions

Security

  • While the sandbox is isolated, exercise caution when handling sensitive data.
  • Do not hardcode API keys or passwords in your code.

Session Management

  • The sandbox is reset after 10 minutes of inactivity, deleting all files.
  • Download important results using the export_file tool.

Resource Limitations

  • The sandbox uses limited CPU and memory.
  • Large-scale data processing or long-running tasks may be restricted.

File Storage Location

  • Always save files to the /workspace directory.
  • Files in other locations may disappear after execution.

Network Restrictions

  • External network access is limited to library installation purposes only.
  • Python: pypi.org, files.pythonhosted.org
  • JavaScript: registry.npmjs.org, nodejs.org
  • Go: proxy.golang.org, sum.golang.org, index.golang.org
  • API calls, web scraping, and other external service access are not possible.

Library Installation

  • Python: Common libraries like pandas, numpy, and requests can be installed
  • Some libraries with system dependencies may not be installable (e.g., matplotlib)
  • JavaScript: npm packages can be installed, but native modules may be restricted
Last updated on
TerminalMySQL