I Wanted Claude Code Running 24/7 on a Server — So I Built VibeCheck

Close your laptop, Claude Code session dies. Anthropic's Remote Control is the same — it's a local process.

"Put it on a server, access from my phone, keep it running even when the laptop is off."

That one sentence is why I built VibeCheck. It's open source (MIT).

Why I Built It

I tried Anthropic's Remote Control. Being able to see a Claude session from your phone is nice, but close the laptop and the session dies. If WiFi drops for more than 10 minutes, it times out. And you only get one session.

It's called "Remote" Control, but your computer has to stay on.

Anthropic also offers a cloud option (claude --remote), but it can't use your local environment — no local MCP servers, no environment variables, no local DB. It clones your GitHub repo and works from there, which doesn't fit projects that depend on local setup.

There was no option that kept the local environment while letting you close the laptop. So I built one.

Architecture

Browser (phone/laptop/tablet)
       |
       v
VibeCheck Server (WebSocket hub)
       |
       v
Your Server (vibecheck-agent + Claude Code CLI)

The agent runs Claude Code headlessly on your server and connects to a relay server via WebSocket. When you open the browser and connect to the relay, you get real-time streaming conversation.

Session history is saved in the same .jsonl format Claude Code CLI uses. Start a session on CLI, continue on the web. Start on the web, continue on CLI.

Installation

Cloud (Simplest)

curl -sL https://vibecheck.sotaaz.com/install/YOUR_API_KEY | bash

One line. Open the browser and start coding.

Self-Hosted (Free)

git clone https://github.com/NestozAI/VibeCheck
cd VibeCheck/self-hosted
./setup.sh

Access at http://localhost:8501. Slack integration is available in the self-hosted version.

Agent Only (npm)

npm i -g vibecheck-agent
vibecheck-agent --key YOUR_API_KEY --dir /path/to/project

Requires Node.js 18+ and Claude Code CLI (claude in PATH).

Key Features

Skill Presets

Seven pre-configured agent modes for different tasks:

Task Scheduler

Cron-based recurring tasks. Daily git pulls, weekly dependency audits, scheduled test runs.

Model Selection

Switch Claude models per query. Haiku for quick questions, Sonnet for coding, Opus for complex reasoning.

Cost Tracking

Token count and USD cost displayed for every response.

Security

Path-based access control by default. Only the working directory is trusted. Access to external paths triggers an approval UI — one-time or permanent.

How I Actually Use It

I use both tools for different situations.

Quick local check-ins — Remote Control. Pull out my phone, ask "did the tests pass?"

Long-running tasks or when I want to close my laptop — VibeCheck. It runs on the server, so the work continues even after I leave.

Demo

Bottom Line

Claude Code is a great tool. But "it dies when you close the laptop" is a bigger constraint than it sounds.

Put it on a server, access from your phone, keep it running with the laptop off — that's why VibeCheck exists.

Cloud: vibecheck.sotaaz.com
Self-hosted: github.com/NestozAI/VibeCheck
License: MIT

References: Anthropic Remote Control Docs · Claude Code on the Web · VibeCheck GitHub