Parallel Development with Claude Code Without git worktree (container-use)

A developer has published a walkthrough of container-use, an MCP tool from Dagger that creates isolated per-branch development environments inside containers, allowing parallel Claude Code sessions without the extra directories that git worktree creates. The setup involves installing Dagger and container-use, registering the MCP server with `claude mcp add container-use -s user -- cu stdio`, and appending the project's agent.md rules to CLAUDE.md so Claude actually invokes the tool. Changes are pulled back into the local repo with `cu checkout <env_id>`, with `cu list` and `cu log` tracking environments and operation history.

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. container-use https://github.com/dagger/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. | 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 | Arch Linuxの場合 sudo pacman -S dagger yay -S container-use For other OSes, see the official repository https://github.com/dagger/container-use . 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 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 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 | 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 | Once development is done, pull the changes from the container into your local repo with: cu checkout <env id Here's a concrete example. Open two terminals and start a separate Claude Code session in each. bash $ claude Issue 42のバグを修正して bash $ 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. 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 The more efficient parallel development gets, the faster review requests pile up. 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. With container-use, you can do parallel development in Claude Code without scattering directories the way git worktree does. Particularly recommended if you keep your repositories nicely organized with ghq. This post was written with AI assistance and reviewed by the author.