Claude Code for Beginners: Set Up an AI Agent in Your Terminal

I've already described in general how AI agents in the terminal change your workflow. Here it gets concrete: how to set up Claude Code as a beginner and run your first real task.

Claude Code is an AI agent for the terminal: it reads and writes files in your project, runs commands, searches code and works through entire tasks – not just a chat snippet. That's exactly the difference from a browser chat: it has hands in your project directory.

The prerequisite is Node.js. Then install globally:

Then change into your project directory and start the agent:

On first start Claude Code walks you through the login – either via your Anthropic account (subscription) or an API key. Important to understand: the agent works in the current directory. So always start it where your project lives – that's its workspace.

Instead of asking for code, you give it a task in natural language – for example:

“Read the README and explain in three sentences what this project does.”

or more concretely:

“Add error handling to the parse_config function and write a test for it.”

Claude Code proposes changes, shows you diffs and asks before critical actions. You stay in control – the agent does the typing.

Claude Code gets its power from tools: read/write files, shell commands, web search. Beyond that it can be connected to your own systems via MCP – what an MCP server is, I explain in detail here. That way the same agent can also drive your server or – as in my case – a whole blog.

Claude Code runs locally but barely needs any resources – in my case on the same power-efficient mini PC that carries the whole homelab:

• CLAUDE.md – project-specific instructions the agent honors automatically; worth it once you work regularly.
• Your own MCP servers & subagents – the next level.
• Permissions/sandbox – fine-tuning what the agent may do without asking.

With two commands your first AI agent runs in the terminal – and the difference from a browser chat is immediately tangible on the first real task. If you want to run the agent permanently on a server and control it from anywhere, it continues here: Claude Code with Remote Control.