Claude Code 如何透過 container-use 在不使用 git worktree 的情況下平行開發

You want to run multiple tasks in parallel with Claude Code, but you'd rather not have git worktree spawn a new directory every time. This post introduces "container-use," a tool that solves exactly that.

Who this article is for

You use Claude Code and are interested in parallel development
You know git worktree, but find its directory management cumbersome
You keep your repositories tidy with ghq or similar tools

What is container-use?

container-use is an MCP tool from Dagger. It creates isolated development environments inside containers on a per-branch basis, letting you do parallel development without polluting your local directory structure.

Comparison with worktree

Aspect	git worktree	container-use
Directories	A new directory is created	Isolated inside a container
Environment isolation	Files only	The whole environment
Work log	Only git history	Operation history via `cu log`
Compatibility with ghq	Tends to get messy	No impact

Setup

1. Installation

# Arch Linuxの場合
sudo pacman -S dagger
yay -S container-use

For other OSes, see the official repository.

2. Add the MCP to Claude Code

# グローバルにMCPを追加
claude mcp add container-use -s user -- cu stdio

# 接続確認
claude mcp list

If you see the following, you're good to go.

container-use: cu stdio - ✓ Connected

3. Add the rules to CLAUDE.md (important)

The official docs mark this as "optional," but without it Claude won't actually use container-use.

curl https://raw.githubusercontent.com/dagger/container-use/main/rules/agent.md >> CLAUDE.md

4. Allow the tools (optional)

If you'd rather not get a confirmation prompt every time, launch with the following options.

claude --allowedTools mcp__container-use__environment_checkpoint,mcp__container-use__environment_create,mcp__container-use__environment_add_service,mcp__container-use__environment_file_delete,mcp__container-use__environment_file_list,mcp__container-use__environment_file_read,mcp__container-use__environment_file_write,mcp__container-use__environment_open,mcp__container-use__environment_run_cmd,mcp__container-use__environment_update

Basic usage

cu command reference

Command	Description
`cu list`	List environments
`cu log <env_id>`	View the work log
`cu checkout <env_id>`	Apply changes to a local branch
`cu open <env_id>`	Open an existing environment
`cu config`	Change settings like the base image

Pulling work back in

Once development is done, pull the changes from the container into your local repo with:

cu checkout <env_id>

Parallel development workflow

Here's a concrete example.

Example: running two tasks at once

Open two terminals and start a separate Claude Code session in each.

Terminal 1 (bug fix)

$ claude
> Issue #42のバグを修正して

Terminal 2 (new feature)

$ claude
> Issue #58の機能を実装して

Each Claude works inside its own container-use environment, so they don't step on each other.

Flow after work is done

1. Claudeが作業完了を報告
   ↓
2. cu list で環境IDを確認
   ↓
3. cu checkout <env_id> で変更を取り込み
   ↓
4. 通常通りレビュー → マージ

Things to watch out for

Review becomes the bottleneck

The more efficient parallel development gets, the faster review requests pile up.

Countermeasures

Keep issues small enough to be easy to review
Ask Claude Code to self-review and generate a diff summary
Set a rule to review in order of priority

Don't forget the CLAUDE.md addition

As noted above, if you don't append the contents of agent.md to CLAUDE.md, Claude won't recognize container-use. Keep in mind that this has to be set up per project.

Summary

With container-use, you can do parallel development in Claude Code without scattering directories the way git worktree does.

Doesn't pollute your directory structure
Environments are fully isolated
Work logs are kept, which is handy for verification

Particularly recommended if you keep your repositories nicely organized with ghq.

References

This post was written with AI assistance and reviewed by the author.