Container Use

What the dagger/container-use MCP server does

The dagger/container-use MCP server provides containerized development environments for coding agents. It is delivered as a command-line MCP server and is powered by Dagger. Agents can work in separate environments and branches, allowing several tasks or experiments to proceed without modifying the same working state.

Each agent environment starts fresh and is associated with its own git branch. Developers can review an agent’s changes through normal git operations, including checking out the relevant branch. Failed experiments can be discarded without affecting the main working environment.

How it works

The server exposes container-use through the stdio command, which an MCP-compatible agent launches as a local process. The agent asks Container Use to create and operate in an isolated environment, while Dagger supplies the container execution layer.

The workflow also records command history and logs. Developers can inspect what an agent actually ran instead of relying only on its report. When an agent is blocked or needs supervision, the developer can open its terminal, inspect the current state, and take direct control.

The dagger/container-use MCP server is intended for parallel coding work, exploratory changes, and tasks where isolation between agents matters. It is not presented as a hosted service or as a replacement for the project’s existing git workflow.

Setup and configuration

Install on macOS with Homebrew:

brew install dagger/tap/container-use

An install script is also provided for all platforms:

curl -fsSL https://raw.githubusercontent.com/dagger/container-use/main/install.sh | bash

Configure the server by adding container-use stdio to the MCP client. For Claude Code, run the command from the repository directory:

claude mcp add container-use -- container-use stdio

The command is also available as cu, so cu stdio can be used as a shortcut. Optional agent rules can be appended to CLAUDE.md. The project links to a setup guide covering Cursor, Goose, VS Code, and other MCP-compatible agents.

Tools and capabilities

The documented capabilities include:

• Creating fresh, isolated container environments for agents.
• Assigning work to separate git branches.
• Running multiple agents without shared-environment conflicts.
• Showing command history and execution logs.
• Providing terminal access for inspection and intervention.
• Supporting review through ordinary git branch operations.

The server is described as compatible with different agents, models, and infrastructure rather than being tied to one model provider.

Limitations and notes

The project is marked as experimental and is still actively evolving. Its documentation describes a CLI-based MCP setup, so the client must be able to launch an MCP server over standard input and output.

The dagger/container-use MCP server should therefore be evaluated as a local development workflow for agent-managed containers and branches. Consult the project’s current documentation before relying on details beyond the installation and stdio setup described here.