How to Install and Use Claude Code: A Beginner's Guide
A step-by-step guide to installing Claude Code, picking modes and models, setting up memory files, and using skills and sub-agents.
What is Claude Code and how do you install it?
Claude Code is Anthropic’s coding-focused AI tool that can read, write, and modify files directly on your computer, not just chat about them. You install it by going to claude.com/download, grabbing the Mac or Windows app, creating an account, and upgrading to a paid plan, since the Code functionality isn’t available on the free tier. Once installed, you point it at an empty folder on your desktop, and that folder becomes its workspace for creating and editing files.
This distinction matters because the regular Claude chat interface behaves a lot like ChatGPT: it answers questions and generates text but can’t touch your file system. Claude Code is a different mode entirely. It can create websites, edit code, organize folders, and run scripts on your machine, which is why it’s useful for actual building rather than just conversation.
TL;DR
- Claude Code runs locally on a folder you create, and it reads, writes, and edits files inside that folder based on plain-English instructions.
- The app costs around $20 a month once you move past the free chat tier, since file-editing capability requires a paid plan.
- Three working modes control autonomy: manual asks permission for every change, plan researches and proposes a roadmap before acting, and auto works independently while a second AI watches for risky behavior.
- Model choice is a tradeoff between cost and power: Haiku is cheapest, Sonnet is the middle ground, and Opus is the most capable but most expensive, with an effort setting that controls how long it thinks before responding.
- Memory lives in a claude.md file, created automatically by typing
/init, which lets Claude retain context about your project across sessions. - Skills turn one-off tasks into reusable workflows, saved as skill.md files that Claude can reference the next time you want it to repeat a job.
- Sub-agents let you parallelize work, handing off side tasks to secondary Claude instances so multiple things get built at once instead of one at a time.
How do you give Claude Code your first instruction?
Once you’ve opened your project folder inside Claude Code (click “local,” then “open folder,” then select the folder you made), you type what you want in plain English in the prompt box and hit enter. The practical advice here is to describe the outcome, not the steps. Instead of walking Claude through a technical sequence, you state the result you want, like “make me a simple website for a cleaning company with a booking form,” and let it figure out the implementation.
This outcome-first approach extends to more complex projects too. You can hand Claude source material, like a floor plan and listing photos, and ask for something as specific as a 3D visualization of an apartment, and it will work out the intermediate steps on its own, including, in one demonstrated case, sourcing furniture products online to populate the space.
If a task starts going sideways, you have two options: hit the stop button to halt it immediately, or type a correction mid-stream and push it through. You don’t have to let a bad direction run to completion before intervening.
What do the different modes and models actually control?
The mode selector, located under the prompt box, determines how much Claude does on its own before checking in with you:
- Manual asks for approval before every single change, giving you maximum control but requiring constant attention.
- Plan has Claude research the task, build a detailed roadmap, and wait for your approval before executing anything.
- Auto lets Claude work mostly independently, with a second AI monitoring in the background to catch risky actions, like deleting files, before they cause damage.
Auto mode is generally the recommended default for most tasks, since the safety layer reduces the risk that comes with full autonomy.
Separately, the model selector picks which underlying AI does the actual work. The lineup runs from Haiku (cheapest, least capable) to Sonnet (a middle-ground option) to Opus (most expensive, most powerful). There’s also an effort setting that controls how long the model thinks before producing an answer, with medium or high generally producing better results at the cost of more usage. Both the model and effort settings directly affect how much of your usage allowance you burn through, since every account has a usage cap before access gets temporarily cut off.
Why does Claude Code need a memory file?
Every interaction with Claude Code happens inside a “session,” and a new session has no memory of what happened in a previous one unless you explicitly give it something to read. The standard solution is a file called claude.md, which stores project context that Claude loads at the start of each session. You don’t have to write this by hand: typing /init tells Claude to scan your workspace and generate a claude.md automatically based on what you’ve already built.
● Reading src/server.ts
● Editing 3 files
+ import { limit } from 'rate'
- app.use(cors())
✓ Tests passed · ready for PR
This matters for a second reason beyond memory: everything in an active session counts as “context,” and Claude has to reread the entire conversation history every time you send a new message. Long sessions get slower, more expensive, and often produce lower-quality output as the context grows. The practical fix is to start a new session whenever you switch topics or begin a new project, rather than letting one conversation balloon indefinitely.
While Claude works, you’ll see line-count indicators (plus and minus signs) showing additions and deletions to your files. You don’t need to parse these to use the tool effectively, but they’re there if you want visibility into what’s changing under the hood. For visual projects like websites or apps, a browser pane on the right side shows you a live preview, which you can use to give direct feedback and request changes.
How do skills and sub-agents extend what Claude Code can do?
Once you’ve built something you’re happy with, you can turn the underlying process into a “skill,” a saved set of instructions stored in a skill.md file that Claude can reuse for similar future tasks. The easiest way to create one is to simply complete the task once, then tell Claude to turn what it just did into a skill. Repeated workflows, like editing videos, drafting client proposals, publishing a podcast, or generating visualizations, become one-step actions instead of tasks you have to re-explain each time.
Claude Code also supports connectors, added via the plus button, which let it interact with outside tools like GitHub, Slack, Linear, Asana, or Trello. Custom plugins can further change how it generates answers.
For scaling up output, Claude Code can run multiple instances simultaneously through sub-agents, which handle side tasks while the main thread continues its own work. This parallelization is what makes it possible to produce many pieces of output (dozens of websites, for example) in roughly the time it would take to build one sequentially.
Finally, the “scheduled” feature lets you queue tasks that run without your involvement, and a mobile app enables remote monitoring and chatting with Claude from your phone. When a task finishes, requesting an “artifact,” a simple static web page summarizing the work, gives you something shareable with clients or colleagues.
Frequently Asked Questions
Do I need to know how to code to use Claude Code?
No. Claude Code is designed to accept plain-English instructions describing the outcome you want, and it handles the technical implementation. Familiarity with code helps for reviewing changes, but it isn’t required to use the tool productively.
What’s the difference between Claude chat and Claude Code?
Claude chat is a conversational interface that cannot modify your files, similar to a standard chatbot. Claude Code is a separate mode that can directly create, edit, and manage files inside a folder on your computer.
Which mode should beginners use, manual or auto?
Auto mode is generally recommended for most tasks because it lets Claude work independently while a secondary AI monitors for risky actions. Manual mode is useful if you want to approve every single change, which can be slower but offers more control for sensitive tasks.
How do I stop Claude Code if it’s doing something wrong?
You can click the stop button to halt it immediately, or type a correction while it’s working to redirect the task without waiting for it to finish.
What is a skill in Claude Code?
A skill is a saved, reusable set of instructions stored in a skill.md file, created by completing a task once and asking Claude to convert that process into a skill you can trigger again later.
