Claude Code builds a phone-accessible dev VM
mgranda reports using Claude Code to create a private development VM, then running six parallel agents from a phone with persistent sessions and notifications when input is needed.
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Create a disposable development VM using the approved cloud account. Configure private network access, persistent terminal sessions, separate Git worktrees and notifications when Claude Code needs input. Document startup, shutdown, authentication limitations and the credentials available to the VM.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- mgranda
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author reports a working remote setup built in one Claude Code session and used to operate six agents from a phone.
- Human involvement
- The owner provides credentials, starts and stops the VM, initiates work and responds to questions; the VM is intentionally disposable.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Supply approved Vultr API access and GitHub CLI access to Claude Code.
- Ask Claude Code to build the private development VM and lifecycle scripts.
- Run separate feature worktrees and agent sessions through tmux from a phone.
The author publishing as mgranda describes a development setup operated from a phone. Claude Code runs on a Vultr virtual machine, with each parallel feature in its own Git worktree and tmux window. Termius provides mobile terminal access, while mosh preserves the connection as the phone changes networks or sleeps.
The concrete setup task was itself delegated to Claude Code. The author supplied a Vultr API key and access to the GitHub CLI, then asked for a secure development VM. They report building the setup in one Claude Code session. The published configuration uses Tailscale-only access, firewall rules and scripts to start and stop the VM.
A notification hook completes the working loop. When Claude asks a question, a script sends the question to a Poke webhook. The phone receives a notification, and the author returns to answer it. This allows a task to continue while the phone is pocketed, with human input when required.
The source reports six parallel agents and branch-based port allocation, but does not give a particular completed feature or code-quality comparison. Its evidenced artifact is the remote development environment and the documented handoff mechanism. The VM runs with permissive agent access and is described as disposable, without production access or unnecessary secrets. The published hourly cost is a historical configuration, not a price guarantee.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Blog · mgranda
Frequently asked questions
What is the completed artifact in this account?
The author reports building the private development VM and mobile access setup in one Claude Code session.
How does the owner know when input is needed?
An AskUserQuestion hook sends the question through a Poke webhook, producing a phone notification.
Does mosh handle every Git authentication need?
No. The source notes that mosh does not forward the SSH agent, so the author uses regular SSH inside tmux when needed.
More details
Remaining steps
- Receive a Poke notification when the agent needs input, then reconnect and respond.
About this account
Claude Code built the development setup from a goal, and running agents continue tasks remotely until a question requires the owner.
Firsthand source opened over HTTP and read on 2026-10-15. Publication date comes from the source page. Outcomes are author-reported and were not independently reproduced. Suggested prompt written by Agents In Use. Evidence supports environment creation and operation; no particular completed feature or code-quality benchmark reported.
Credit: mgranda
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Vultr · Tailscale · Termius · mosh · tmux · Git worktrees · Poke
Mobile development · Remote workspace · Human handoffs
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.