Agent Sandboxes

Here we explore e2b agent sandboxes and claude code together to scale your agentic engineering.

Watch the full video where we break down agent sandboxes and claude code here

Using Agent Sandboxes (E2B) for complete agentic engineering control.

Value Proposition

Agent Sandboxes unlock 3 key capabilities for your agentic engineering:

• Isolation: Each agent fork runs in a fully isolated, gated E2B sandbox, this means no matter what your agent does, it's secure and safe from your local filesystem and production environment.
• Scale: You can run as many agent forks as you want, each fork is independent and has its own sandbox. This is a very literal way to scale your compute to scale your impact.
• Agency: Your agents have full control over the sandbox environment, they can install packages, modify files, run commands, etc. This means they can handle more of the engineering process for you.

Apps

• sandbox_workflows/ - obox: Run parallel agent forks in isolated E2B sandboxes for experimentation
• sandbox_mcp/ - MCP server wrapping sandbox_cli for LLM integration (works from root)
• sandbox_cli/ - Click CLI for E2B sandbox management (init, exec, files, lifecycle)
• sandbox_fundamentals/ - E2B SDK learning examples and patterns
• cc_in_sandbox/ - Run Claude Code agent inside an E2B sandbox (ibox: in box agent)
• sandbox_agent_working_dir/ - Agent runtime working directory

Agent Sandbox Tooling Choice

Using e2b (General Sandbox SDK) for:

• Full control over sandbox environment
• Shell command execution
• File system operations
• Running tools (Claude Code, git, npm, etc.)

Quick Start

1. Global Environment Setup

Create a .env file in the project root with the following API keys:

# Required for all sandbox operations
E2B_API_KEY=your_e2b_api_key_here
ANTHROPIC_API_KEY=your_anthropic_api_key_here

# Optional: Required for git push/PR functionality
GITHUB_TOKEN=your_github_token_here

Install the top Agentic Coding Tool:

• Claude Code: https://www.claude.com/product/claude-code

Get your API keys:

• E2B API Key: https://e2b.dev/docs - Sign up and get your API key
• Anthropic API Key: https://console.anthropic.com/ - Create an API key in your account settings
• GitHub Token: https://docs.github.com/en/authentication/keeping-your-account-and-data-secure/managing-your-personal-access-tokens - Create a personal access token with repo scope

2. Application Usage

I recommend working from the atoms of the codebase (e2b fundamentals, cli, mcp) to the full e2e sandbox workflow (obox workflows).

Process:

1. Setup environment variables
2. Explore E2B Fundamentals apps/sandbox_fundamentals/
3. Explore E2B CLI apps/sandbox_cli/
4. Explore E2B MCP apps/sandbox_mcp/
5. Explore E2B Workflow apps/sandbox_workflows/

Explore E2B Fundamentals - apps/sandbox_fundamentals/

Start Here

Recommended: Walk through all example scripts to understand E2B sandbox concepts.

cd apps/sandbox_fundamentals
uv sync

# Run through all examples in order
uv run python 01_basic_sandbox.py
uv run python 01_basic_sandbox_keep_alive.py
uv run python 02_list_files.py
uv run python 03_file_operations.py
uv run python 04_run_commands.py
uv run python 05_environment_vars.py
uv run python 06_background_commands.py
uv run python 07_reuse_sandbox.py
uv run python 08_pause_resume.py
uv run python 09_claude_code_agent.py
uv run python 10_install_packages.py
uv run python 11_git_operations.py
uv run python 12_custom_template_build.py
uv run python 12_custom_template_reuse.py
uv run python 13_expose_simple_webserver.py
uv run python 13_expose_vite_vue_webserver.py

Use CLI for Sandbox Management - apps/sandbox_cli/

cd apps/sandbox_cli
uv sync

# Get help
uv run python src/main.py --help

# Initialize a new sandbox
uv run python src/main.py init

# Create a sandbox with custom template (this is an e2b sandbox template - you can create this by running the `uv run python 12_custom_template_build.py` script)
uv run python src/main.py sandbox create --template agent-sandbox-dev-node22

# Execute a command in a sandbox
uv run python src/main.py exec <sandbox-id> "ls -la"

# List files in a sandbox
uv run python src/main.py files ls <sandbox-id> /

You can also boot up a claude code agent and run /prime_cli_sandbox.md - then prompt your agent to run the commands for you.

Use MCP Server with Claude Desktop - apps/sandbox_mcp/

Works from project root - MCP server is configured in your Claude Desktop config.

# cp the .mcp.json.sandbox to .mcp.json
cp .mcp.json.sandbox .mcp.json

# replace your e2b api key in the .mcp.json env section
...

# Start a claude code agent with the .mcp.json
claude

# Check the mcp server status
/mcp

# Prompt the same commands as you would with the sandbox_cli with natural language
prompt: What can we do with the e2b sandbox tools?

prompt: init a new sandbox

prompt: create a sandbox with custom template agent-sandbox-dev-node22

prompt: run ls -la in the sandbox

prompt: search for all .py files in the sandbox with exec

# Run custom slash commands
prompt: /plan Add buttons to the nav bar that auto scroll to respective sections on the landing page

prompt: /build <path-to-plan>

prompt: /wf_plan_build Add buttons to the nav bar that auto scroll to respective sections on the landing page

Run Parallel Agent Experiments - obox - apps/sandbox_workflows/

cp .mcp.json apps/sandbox_agent_working_dir/.mcp.json (after you fill it out with your e2b api key)
cp .env apps/sandbox_agent_working_dir/.env (after you fill it out with your credentials)

cd apps/sandbox_workflows
uv sync
uv run obox <repo-url> --branch <branch> --model <opus|sonnet|haiku> --prompt "your task" --forks 3

You can also boot up a claude code agent and run /prime_obox.md - then prompt your agent to run the commands for you.

See apps/*/README.md for detailed documentation on each tool.

Application Notes

obox - apps/sandbox_workflows/

• There are two system prompts for the custom agent that drives the sandbox engineering
  • apps/sandbox_workflows/src/prompts/sandbox_fork_agent_w_github_token_system_prompt.md - Supports GitHub token auth for git push/PR functionality
  • apps/sandbox_workflows/src/prompts/sandbox_fork_agent_system_prompt.md - Basic system prompt for sandbox engineering, does not have git push/PR functionality
  • You can either use specific prompt workflows (see apps/sandbox_agent_working_dir/.claude/commands/wf_plan_build.md) to manage git operations, or you can use the system prompt.
• The agents working directory is apps/sandbox_agent_working_dir/ see apps/sandbox_workflows/src/modules/constants.py for more details.
  • That means that apps/sandbox_agent_working_dir/./claude/commands/ are the available slash commands for the agent (and all the other claude capabilities are active there too)

Resources

• https://e2b.dev/
• https://www.claude.com/product/claude-code
• https://docs.claude.com/en/docs/agent-sdk/python
• See ai_docs/README.md for resources used to build this codebase.

Master Agentic Coding

Prepare for the future of software engineering

Learn tactical agentic coding patterns with Tactical Agentic Coding

Follow the IndyDevDan YouTube channel to improve your agentic coding advantage.