How do I run Claude Code inside the Muse VM?

Ask your Muse to download and install Claude Code in its VM, then complete Claude Code’s sign-in with your own Claude subscription so Muse can drive the command-line tool for you. Keystroke did this in October 2026, then had Muse write a skill for how to use Claude Code and a CLAUDE.md file that explains his Muse setup to Claude.

The Muse VM is the cloud machine where your Muse runs commands and scripts. Claude Code is Anthropic’s coding agent that runs from a terminal, so Muse can run it like any other program there.

Steps

Keystroke described his setup in the Muse Agent server’s #general channel on October 20 and 22, 2026:

1. Have Muse install it. He asked his Muse to download and install Claude Code in the VM.
2. Sign in. He completed the OAuth sign-in so Claude Code uses his Claude subscription, one he set up just for this.
3. Write a skill for it. Muse’s first task with Claude Code was to set it up and build a claude-code skill, which tells Muse when and how to use it for writing code.
4. Curate a CLAUDE.md. CLAUDE.md is the file Claude Code reads for project context. Keystroke’s describes the parts of the Muse system that matter for his tasks and the VM-wide git repository Muse keeps.
5. Let Muse drive. “In general, let your muse be the one driving it,” he advised. He treats Claude Code as a tool his Muse uses, not a separate assistant.

Make it watchable

Keystroke’s further tips, for anyone setting this up:

• Ask Muse to build a dashboard artifact (a small web page Muse makes and hosts), and to call Claude Code through a helper script that tees Claude Code’s streaming JSON output into a folder. The dashboard can then follow all Claude Code activity as it happens. He posted a short video of his.
• Tell Muse that those responses contain metadata on token usage and remaining balances, and have the skill report spend and usage back to you as it goes.
• His Muse calls Claude Code through subagents, so you can keep chatting with Muse while it runs one or several Claude Code sessions in the background. That is where the dashboard helps most.

He and his Muse settled on what he called a builder pattern, similar to how Muse works with its own artifact builder. “There is a lot of personal taste involved,” he added, “so feel free to experiment.”

Samm, who had just bought a separate plan for her Muse, Willow, asked for tips on October 22; the list above was his reply.

A different setup: Claude outside the VM

Chonky runs it the other way round. His Claude Code agent runs on his own computer, and he puts it and his Muse in a Discord server where they talk to each other: Claude does the coding, and Muse does image and video generation and DevOps work. When Keystroke asked why he used Discord as the go-between, Chonky said he liked seeing the bots work as a team, being able to read what they say to each other, and being able to step in with directions, while admitting it might not be the most efficient arrangement.

Why it works

Keystroke’s reply to Chonky sums up the difference: in his setup Muse directly invokes Claude Code, with no intermediary, because the tool lives on the same machine as Muse. Muse already runs code in its VM; Claude Code is one more program there, and the skill and CLAUDE.md give each side the context it needs about the other.