Most people use AI in a chat window. Write me an email. Summarize this document. That’s fine, but it’s like having a carpenter and only asking them to hand you tools.

I’ve used Claude Code for about five months. It’s Anthropic’s command-line tool, and it can read files, run shell commands and call APIs directly. What changed everything for me wasn’t a better prompt or a smarter model. It was giving it access to real infrastructure.

The stack

The whole setup costs under $10 a month:

  • A VPS from Hostinger (~$7/mo). A small Linux server with 2 vCPUs and 8GB RAM.
  • Coolify (free, open source). It manages Docker containers, SSL certificates and domain routing. Think of it as your own little Heroku.
  • N8N (free, self-hosted). It’s like Zapier or Make, but you own it, it runs on your server, and there are no per-task limits.

I bought the VPS and pointed it at a domain. Claude Code did the rest: it SSHed in, installed Coolify, deployed N8N and configured the DNS routing. Now it keeps working with that setup through the Coolify API and an MCP server that connects to each N8N instance.

Give it its own limited user, not root, and keep backups before you let it change anything.

Automations it builds and tests

I can say, “Build me a workflow that watches a webhook, fetches the page content, and saves it to my Notion database.” Claude Code looks up how each N8N node works, writes the workflow, deploys it live, runs a test, reads the log and fixes whatever broke.

Once that workflow is running, there’s no AI in the loop. It fires on a schedule or a webhook and costs nothing per run.

Deploys it checks

I run about a dozen small services across a couple of servers, each one a Docker container managed by Coolify. When I deploy one, Claude Code builds the image and brings the new version up. Then it hits a /version endpoint on the running service to confirm the deploy actually worked.

That check exists because deploy tools can lie to you. For a stretch, my deploy API returned success while quietly restarting the old container. Seven of nine services were affected, and I didn’t notice for weeks. Claude Code found the problem by querying the Coolify API across all my servers. Then it rebuilt the deploy pipeline with SHA-based version checks in every app.

Use AI surgically

Sometimes a small AI step is worth it. The word to stress is small.

My bookmark system is an eight-node N8N workflow. I tap Share in Safari, and it fetches the page, cleans the HTML and saves it to Notion. Seven nodes are fully deterministic. One calls Claude Haiku, the smallest, cheapest model, to clean up the title, write a one-sentence description and pick 1-3 tags from my existing 58.

The workflow doesn’t trust the AI blindly. If Haiku returns a tag that doesn’t exist in my database, the next node drops it. If the AI call fails entirely, the bookmark still saves, just without the extras.

The less AI it needs later, the better

When AI can SSH into a server, call APIs, and build and test workflows, it stops being a conversation. It builds the thing, deploys it, checks it works and fixes what’s broken. Then the thing just runs on your server, with no ongoing API costs and no dependency on whether the AI service is up tomorrow.

AI is good at building things. But the best systems it builds for me are the ones that don’t need it anymore.

A $7 server and some open-source tools got me there. I think it’d work for a lot of people.