[Feature]: create new environment manually?

[vitorbal]

What feature would you like?

Would it be possible to add a new command to container-use so I can create an environment by hand, the same way an agent can via the environment_create tool?

Why would this be useful?

I would like to use container-use for a workflow where I:

1. create an environment by hand: (e.g. a new container-use create <env> command)
2. get a shell for that environment
3. Call Claude Code from inside that environment
4. Let Claude Code operate within the environment with its native tools, instead of container-use tools

I'm not interested in using container-use via an agent/MCP server.

[cwlbraa]

this is a common request, and we're cooking some stuff in this department. :)

[mwotton]

really feels like this should be the main usage mode. the whole point of using an environment is to isolate changes - i don't want to have to rely on the LLM to decide to put itself in gaol.

[cwlbraa]

@mwotton yeah, strongly agreed. the way i've been using container-use, I configure the "agent" (claude code, zed, cursor, whatever) to put the LLM in container-use jail --- that's better than relying on the LLM to stay in it's own imaginary prompt engineering jail, but it's still not exactly what the people want.

the hard part is that having container-use jail the agent fundamentally changes the interface between container-use and the agent: to do this, container-use can no longer be "just an MCP server"