Claude Code for beginners
Most AI coding help still looks like a chat window: you paste code in and copy an answer out. Claude Code works differently. You start it inside your project, and it reads the files, makes edits, runs commands and checks the results. Anthropic calls it an "agentic" coding tool. In plain English: an assistant that can act, not just answer.
This guide starts our Claude Code series. Every price, limit and command below comes from Anthropic's own docs and pricing pages, checked on October 20, 2026.
01What Claude Code actually is
Claude Code works in a loop: look at the problem, take an action, check what happened, decide the next step, until the job is done or it needs you. It started in the terminal, and the official docs now list five places to run it:
- Terminal. The full-featured command-line version.
- VS Code and Cursor. An extension with inline diffs (side-by-side views of what changed).
- JetBrains IDEs. A plugin that needs the terminal version installed too.
- Desktop app. Visual diff review and parallel sessions. Needs a paid plan.
- Web. At claude.ai/code, running in the cloud with no local setup.
One thing surprises people: Claude Code isn't open source. Its GitHub repo holds the changelog, issue tracker and example plugins, but the license file reads "© Anthropic PBC. All rights reserved."
02Install it in two minutes
You need macOS 13+, Windows 10 (1809) or newer, Ubuntu 20.04+, Debian 10+ or Alpine 3.19+, plus 4 GB of RAM, per the setup docs. On macOS, Linux or WSL (Windows' built-in Linux layer), open a terminal and run:
curl -fsSL https://claude.ai/install.sh | bashOn Windows PowerShell, use irm https://claude.ai/install.ps1 | iex. The installer shows no progress while it downloads. When it's done, open a new terminal window and check:
claude --versionThe native installer updates itself. Homebrew and WinGet installs don't, and the README marks the old npm install as deprecated. Never used a terminal? Start in the desktop app or VS Code instead.
03Sign in: which plan do you need?
The first time you run claude, it asks you to log in. The docs recommend a Claude subscription: a flat monthly price with usage limits. The alternative is a Claude Console account, which is pay-as-you-go API billing from prepaid credits. The free plan doesn't include Claude Code.
Chart 1
Individual Claude plans, monthly price
| Plan | Price per month | Usage |
|---|---|---|
| Free no Claude Code | Chat only. Claude Code not included. | |
| Pro the starting point | At least 5x Free per five-hour session | |
| Max 5x heavy daily use | 5x Pro per five-hour session | |
| Max 20x all day, every day | 20x Pro per five-hour session |
Usage limits and paying per token
Usage resets on a rolling five-hour window, and paid plans add weekly limits on top. Your Claude chats draw from the same pool as Claude Code, and Anthropic publishes no fixed message count. Hit a limit and you can wait, upgrade, or turn on usage credits billed at API rates.
On the API route you pay per token (roughly a chunk of a word). The pricing page listed Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens. Anthropic's cost docs say enterprise deployments average about $13 per developer per active day: Anthropic's figure for company teams, not a forecast for you. A sensible default is to start on Pro and move up only when you keep hitting the limit.
04Your first hour
Chart 2
Your first hour, step by step
claude --versionInstall and confirm it runs.claudeStart it in a project folder and log in.what does this project do?Ask questions before changing anything./initGenerate a starter CLAUDE.md.commit my changesMake one small change, review it, commit it.
Start Claude Code inside a project folder, ideally a side project or a copy:
cd your-project
claudeThen just ask, for example "what does this project do?". Claude reads files as it needs them, so you don't paste anything in. Next, run /init: it scans the project and writes a starter CLAUDE.md with the build commands, test steps and conventions it finds.
Now try one small change, such as "add input validation to the signup form". Claude shows the edit, you approve it, then ask it to "commit my changes with a descriptive message". Three keys to learn now: Shift+Tab cycles permission modes, Esc twice opens the undo menu, and /help lists every command.
05Five ideas that make it click
CLAUDE.md: the project's memory
A plain markdown file in your project root that Claude reads at the start of every session. Put rules there like "use pnpm, not npm". The docs suggest keeping it under 200 lines, because longer files get followed less reliably. Claude also keeps its own auto memory, which you can review with /memory.
Permission modes: how often it asks
Chart 3
What each permission mode lets Claude do without asking
| Mode | Edits files | Runs commands | Docs say it's for |
|---|---|---|---|
Manual (default) | No | Partly read-only ones | Reviewing every action, sensitive work |
acceptEdits | Yes | Partly common file commands | Iterating on code you're reviewing |
plan | No until you approve | Partly classifier-approved | Exploring before changing anything |
auto | Yes | Yes with safety checks | Long tasks, fewer prompts |
bypassPermissions | Yes | Yes no checks | Isolated containers and VMs only |
dontAsk, is meant for locked-down scripts and CI. Deny rules you set apply in every mode.A big change landed in August 2026: auto mode became the default starting mode for new sessions, first on Pro, Max and Team plans, and from version 2.1.283 on every plan. A separate classifier model checks each action and blocks what it judges unsafe, but the docs say auto mode "does not guarantee safety". On a project that matters, start with claude --permission-mode default so it asks first.
Plan mode: think before typing
Press Shift+Tab until plan mode is on, or start a prompt with /plan. Claude researches and writes a plan but can't edit your code until you approve it. Use it for anything bigger than a one-file fix.
Checkpoints: an undo button
Every prompt creates a checkpoint. /rewind, or Esc twice on an empty prompt, rolls back the code, the conversation or both. It isn't a replacement for git: the docs note that edits made by subagents usually aren't captured.
Context: its working memory
Every message, file and command output fills the "context window", the text the model can hold at once, and Anthropic says performance drops as it fills. /clear starts fresh and /compact summarises.
06Subagents, skills, hooks, MCP and plugins
- Subagents are helpers with their own context window that hand back a summary, such as the built-in Explore agent for searching code.
- Skills are
SKILL.mdinstruction files that become a slash command like/deploy, or that Claude loads when relevant. - Hooks are shell commands that run automatically at set moments, such as formatting code after every edit.
- MCP (Model Context Protocol) is an open standard for connecting Claude to outside tools like Notion, Jira or your database.
- Plugins bundle all of the above into one install. Anthropic's official marketplace is added for you.
The first line below runs in your terminal, the second inside a session:
claude mcp add --transport http notion https://mcp.notion.com/mcp
/plugin install commit-commands@claude-plugins-officialChart 4
Which feature solves which problem
| You notice that… | Reach for | Day one? |
|---|---|---|
| Claude gets a convention wrong twice | A line in CLAUDE.md | Yes |
| You keep typing the same prompt to start a task | A skill | Partly |
| A side task floods the chat with output | A subagent | Partly |
| You keep copying data from a tool Claude can't see | An MCP server | No |
| Something should happen every time, without asking | A hook | No |
| A second project needs the same setup | A plugin | No |
07Good habits for founders
- Use git every time. Commit or branch before each task, so any change is one command from undone.
- Keep tasks small and specific. The quickstart's example: "fix the login bug where users see a blank screen after entering wrong credentials" beats "fix the bug".
- Give it a way to check its work. Tests, a build or a linter. The best-practices guide puts it simply: "If you can't verify it, don't ship it."
- Read the diff.
/diffshows what changed, and/code-reviewand/security-reviewcheck it. Per the security docs, reviewing it is your job. - Reset when it goes in circles. After two failed corrections, the docs recommend
/clearand a better first prompt.
08The catches
- Costs can creep. Limits are shared with your chats, and extra usage bills at API rates. If an
ANTHROPIC_API_KEYis set on your machine and you approve it, Claude Code uses that key instead of your subscription. Run/statusto see which one is active. - Auto mode isn't a safety net. It cuts prompts, not risk. Keep anything sensitive in Manual mode.
- Plugins and MCP servers run code. The docs warn that a plugin "can execute arbitrary code on your machine with your user privileges", and that Anthropic doesn't control third-party plugins.
- Prompt injection. Web pages or files can hide instructions aimed at the AI. Anthropic advises reviewing commands and not piping untrusted content into Claude.
- Your data. On Pro and Max you choose whether your sessions can train future models (5-year retention if yes, 30 days if no). Transcripts also sit in plain text under
~/.claude/projects/for 30 days by default. - Lock-in. It's closed source and runs Claude models only. Your CLAUDE.md and skills are plain markdown, and it can read an
AGENTS.mdfile, so the knowledge travels even if the tool doesn't. - It changes weekly. Features get added and removed (the Ultraplan preview went in August 2026), so older tutorials go stale fast.
09Verdict and next step
Claude Code turns "I can't build this fast enough" into a conversation with something that writes, runs and checks code. Pro at $20 a month is a low-risk way to find out whether it fits how you work. Start in a side project, keep git close and read every diff.
Once the basics feel natural, the next step is making it work like a team. Our oh-my-claudecode guide covers a free plugin that adds planners, reviewers and verification loops, and our Superpowers guide looks at ready-made skills.
SourcesSources and further reading
- Claude Code documentation: overview, quickstart, system requirements and authentication.
- Permission modes, memory, checkpointing and best practices.
- Manage costs, security, plugin security and data usage.
- What's new, August 3–7, 2026: auto mode becomes the default.
- anthropics/claude-code on GitHub: README, license and changelog (version 2.1.303).
- Claude pricing and the Max plan help article.
Facts were checked on October 20, 2026. Claude Code changes quickly; if something here is out of date, tell us on X.
Your Average Tech Founder
Host of The relatable entrepreneurship podcast, recorded in Da Nang, Vietnam. Writes about the tools, repos and habits that help founders in the making build better. No guru energy.