Developer tools#Multi-agent#Coding agent
Container Use: a container and branch per coding agent
Dagger's server gives each coding agent its own container and git branch, so several can work in parallel, with command logs, live takeover and git review.
Project and installation docs
View projecthttps://github.com/dagger/container-use
Run two or three coding agents at once and the trouble is that they share one working directory: one installs dependencies, another runs tests, a third edits config, and when something breaks it’s hard to say who did it. Container Use gives each agent a fresh container tied to its own git branch, so the agent can make a mess in there while you keep working outside. It’s built on Dagger and had about 4k stars as of 2026-10-06.
What it does
- Isolated environments: one fresh container and one git branch per agent, so parallel agents don’t collide and failed experiments get thrown away.
- Real visibility: a full history of the commands each agent actually ran, plus logs, not just what it says it did.
- Direct intervention: drop into an agent’s terminal to inspect its state or take over when it gets stuck.
- Plain git workflow: review the work with
git checkout <branch_name>and decide whether to merge. - No lock-in: works with any MCP-capable agent or model; setup is always adding
container-use stdioas an MCP server.
Who it’s for
- Developers running several Claude Code or Cursor sessions on different tasks without letting them trash the local environment.
- Anyone uneasy about agents installing packages or touching system files on the host.
Setup
On macOS install with Homebrew (other platforms have an install script), then register it from your repository:
brew install dagger/tap/container-use
cd /path/to/repository
claude mcp add container-use -- container-use stdio
The README also offers an optional agent rules file to append to CLAUDE.md, telling the agent to work inside its environment.
Our take
Most coding-agent tooling makes the agent write better code. Container Use makes sure that while it writes, it stays out of your way. That’s a different layer, so it stacks with tools like Serena or the GitHub server. Pairing containers with branches also brings review back to ordinary git. The caveats: the project labels itself experimental and the README says it’s in early, fast-moving development, and because Dagger runs the containers locally, your machine needs a working container environment. Licensed Apache-2.0.