e2b

The E2B Sandbox skill provides developers and AI agents with secure, ephemeral, and isolated virtual machine environments optimized for code execution. Designed to be triggered by Claude Code or other autonomous agents, this skill allows for the safe execution of untrusted or AI-generated code, preventing host system contamination. It supports rapid provisioning, typically starting in under 150ms, and provides full filesystem, network, and process management capabilities. This makes it an ideal tool for coding agents requiring a robust execution environment for testing, debugging, and data processing tasks.

• Multi-language runtime support: Executes Python, JavaScript, TypeScript, R, Java, and Bash natively within isolated environments.

• Sandbox Lifecycle Management: Full programmatic control over sandbox creation, pausing, resuming, connection handling, and termination.

• Advanced File & Git Operations: Allows direct writing, reading, and manipulation of files within the sandbox, alongside native Git support for cloning, committing, and pushing repositories.

• MCP Gateway Integration: Connects to over 200+ tools through the Model Context Protocol (MCP) gateway, enabling agents to interact with third-party services like GitHub, Slack, or cloud storage buckets (S3, GCS, R2).

• Infrastructure & Monitoring: Provides real-time metrics for CPU, memory, and disk usage, as well as support for SSH, PTY access, and custom Build System 2.0 templates.

• Use this skill when an AI agent needs to perform calculations, test code, or run bash commands that require external library dependencies (e.g., pandas, numpy).

• It is highly recommended for agents that need to perform multi-step tasks like cloning a repository, modifying code, running tests, and reporting errors back to the user.

• Ensure the E2B_API_KEY environment variable is configured in your global settings for persistent access.

• Note that Python execution requires the use of the Sandbox.create() method as per SDK v2 standards.

• Ideal for parallel execution by creating distinct code contexts, allowing multiple agents or tasks to run in isolation on the same infrastructure.