Why Run Claude Code Remotely

Claude Code, as Anthropic designed it, runs on a developer workstation. You open a terminal, launch claude, type prompts, and watch the agent act on your local filesystem. Close the terminal, and the session ends. Reopen it, and you start fresh with whatever is in your CLAUDE.md files.

Remote Claude Code flips this model in three ways:

1. Persistence. Instead of Claude dying when you close the terminal, a remote Claude Code runs as a systemd service on a VPS. It is always on. Conversations persist across reboots. The agent can work on tasks overnight and have results ready when you wake up.

2. Remote access. Instead of needing a laptop with your development environment to interact with Claude, you message a private Telegram bot from your phone. The bot relays messages to Claude Code on the VPS. Claude reads the message, thinks about it, calls tools, and replies through the same Telegram channel. You never open a laptop.

3. Autonomy. Instead of sitting at the keyboard approving every tool call, a background watcher script auto-approves permission prompts so the agent runs unattended. Claude can respond to webhooks, CI failures, monitoring alerts, or Telegram messages while you sleep.

The result is an AI coding agent that feels less like a tool you use and more like a teammate who is always online. That is a genuinely different experience from desktop Claude Code, and it carries different risks and rewards.

Not everyone needs remote Claude Code. The operational overhead of running a VPS, managing systemd services, handling Telegram bots, and auditing security is real. Before investing the time, be honest about whether your workflow actually benefits.

Strong fit use cases:

• Always-on automation agent. You want Claude Code responding to webhooks from your CI, your monitoring, your CMS, or your ticketing system without you being at a keyboard.

• Phone-first interaction. You travel, commute, or just do a lot of work from your phone. Typing prompts into a Telegram chat is dramatically faster than SSHing in and using a mobile keyboard on a terminal.

• Long-running tasks. You fire off a large refactor, code review, or analysis job that takes 30+ minutes. You want to walk away and come back to results, not babysit the session.

• Team-shared infrastructure agent. Multiple team members want to invoke Claude Code against a specific project. A central VPS deployment with an allowlisted Telegram bot serves this cleaner than each person running their own local instance.

• Hobbyist-to-tinkerer home lab. You already have a VPS running other services. Adding Claude Code costs effectively nothing in marginal spend.

Poor fit use cases:

• Solo developer at one workstation. If your primary workflow is coding at a single desktop, the local Claude Code experience is simpler, safer, and faster.

• Compliance-heavy environments. If you work with data that cannot leave your corporate network, running Claude Code on a public VPS is a policy violation.

• Infrequent use. If you open Claude Code once a week for 30 minutes, running a 24/7 service is massive overkill.

• Beginner comfort with Linux. If you are not comfortable with systemd, SSH, firewalls, and non-root user management, the time cost of learning all of that for this specific use case is high.

A lot of cloud-agent tutorials omit honest pricing. Here is the real monthly cost of running remote Claude Code on the cheapest defensible setup.

Base infrastructure:

Claude API usage:

Anthropic API costs depend on your model and volume. Rough estimates for a single-user agent handling 30-50 interactions per day:

Total realistic spend:

• Hobbyist (Sonnet, low volume): $28-36/mo
• Power user (Sonnet, high volume): $25-55/mo
• Enterprise (Opus, production): $50-115/mo

These are small numbers compared to hiring a junior engineer, but they are real numbers. Many users forget that API usage scales with how aggressively the agent uses tools. A watcher script that auto-approves everything can burn through tokens fast if you give Claude ambiguous tasks.

The hidden cost: your time. Initial setup is 30-90 minutes if you are comfortable with Linux. Ongoing maintenance (token rotation, security patches, occasional debugging) is 1-3 hours per quarter. Factor that into the decision.

Three deployment models cover most users. Pick the one that matches your workflow today, and upgrade later if needed.

Model A: Desktop-only (the default)

Claude Code runs on your laptop. You interact via terminal. Nothing runs when the laptop is closed. This is where most users should start and stay.

• Cost: $0 beyond your Claude Pro subscription
• Complexity: minimal
• Best for: solo developers, local projects, privacy-sensitive work
• Trade-off: no remote access, no persistence, no automation

Model B: Hybrid (desktop primary + VPS secondary)

Your primary Claude Code runs on the laptop. You also run a smaller agent on a VPS that handles specific async jobs (webhooks, scheduled tasks, phone-triggered queries). The two do not share state.

• Cost: VPS + API = $15-40/mo additional
• Complexity: moderate (managing two Claude Code instances)
• Best for: developers who want async automation but still do heavy work locally
• Trade-off: context is not shared between the two agents

Model C: Full remote (VPS primary)

All Claude Code activity happens on a VPS. You interact via Telegram from your phone, or SSH into the VPS from any machine. No local Claude Code installation needed.

• Cost: VPS + API = $25-55/mo
• Complexity: high (systemd, security, operations, backup)
• Best for: phone-first power users, teams with shared agent needs, infrastructure engineers
• Trade-off: you depend on the VPS being healthy; outage = no Claude Code at all

Our recommendation: start with Model A until you hit a specific limitation, then jump to Model B (not C). Model C is for users who have proven they need always-on automation and are comfortable with VPS operations. The rest of this Pro Track teaches Model C end to end.