Claude Code: What It Is, What It Costs and How to Start
Claude Code explained: what it is, what it costs, how to install it on Mac or Windows and how to start. Practical guide with pricing, a Cursor comparison, FAQ.
Claude Code is an AI coding agent from Anthropic that runs in your terminal (and also in a desktop app and in editors like VS Code). Unlike a chat, it does not just give advice: it reads the files in your project, edits code, runs commands and tests, and shows you the result for approval. You can use it with a Claude Pro subscription (from 17 USD a month billed annually) or higher. The free plan does not include Claude Code.
Claude Code is now my main working environment, whether I am building an app, a website or an automation. This is the guide I would have wanted when I started: what it is, what it costs, how to install it, and which habits decide whether it saves you time or eats it.
What is Claude Code, and how is it different from ChatGPT or Claude in the browser?
The difference is where the AI works and what it can reach.
In a browser chat you paste in a piece of code, get an answer and copy it back into your project by hand. The model only sees what you paste.
Claude Code starts directly inside your project folder. The agent explores the structure on its own, finds the relevant files, proposes changes across several files at once, runs the build or the tests, reads the error message and tries again. Meanwhile you approve individual steps or give it a longer leash.
In practice, a task like "add CSV export to the app and write a test for it" becomes one sentence, and the agent carries it from reading the code to a green test. In a chat you would assemble the same thing piece by piece.
How much does Claude Code cost?
Claude Code is not a separate product. It is part of a Claude subscription. Prices according to claude.com/pricing (October 2026):
| Plan | Price | Claude Code |
|---|---|---|
| Free | 0 USD | no |
| Pro | 20 USD a month, 17 USD billed annually | yes |
| Max | from 100 USD a month | yes, with higher limits |
| Team / Enterprise | per seat | yes |
The alternative is pay-as-you-go through the API (an Anthropic Console account), or through Amazon Bedrock, Google Cloud or Microsoft Foundry. For an individual who wants to try it, Pro is the sensible start. If you keep hitting limits during the day, that is the signal Max will pay off.
How do you install Claude Code?
Supported systems are macOS 13+, Windows 10 (1809+), Ubuntu 20.04+, Debian 10+ and Alpine Linux. You need 4 GB of RAM and an internet connection.
- Open a terminal. Terminal on a Mac, PowerShell on Windows.
- Run the install command. On macOS, Linux and WSL:
curl -fsSL https://claude.ai/install.sh | bash. In Windows PowerShell:irm https://claude.ai/install.ps1 | iex. On a Mac,brew install --cask claude-codealso works. - Check the install. In a new terminal window,
claude --versionprints the version number. If something is off,claude doctortells you what. - Go to your project folder (
cd path/to/project) and runclaude. - Log in. On first run a browser opens and you sign in with your Claude account.
If you would rather avoid the terminal, use the Claude desktop app or the VS Code and JetBrains extensions. The agent is the same, only the window differs.
How do you start your first project?
The best first task is small but real. Not "build me an online store", but "this project has no email validation in the signup form, add it and show me how to test it".
The process that works for me:
- Run
/init. Claude Code goes through the project and creates aCLAUDE.mdfile describing its structure, commands and conventions. It loads this file in every later session, so you do not have to keep explaining how to run the project. - Add the rules you care about to
CLAUDE.md. For example "don't add a new library without asking", "always run the tests after a change", "UI copy is in Czech". Short and concrete. - Start bigger tasks with a plan. Shift+Tab switches to plan mode, where the agent proposes an approach first and changes nothing. Read the plan, correct it, and only then let it work.
- Review the diff, not the conversation. What the agent says matters less than what it actually changed. Look at the changes in git before you commit them.
- Commit often. When a session goes wrong, you want a place to go back to with one command.
What can Claude Code do beyond writing code?
This is where it gets interesting even for people who do not think of themselves as programmers.
It drives your tools through MCP. The Model Context Protocol is an open standard that lets the agent reach external systems: a database, a CMS, task management, Figma, a browser. You add a server with claude mcp add. I went through which ones earned a place and which I deleted in The MCP servers I actually use.
It works with the shell. It can run git, ffmpeg, Python scripts or a deployment. Plenty of "one-off" jobs I used to write a script for are now a single sentence.
It supports skills and custom commands. A repeated procedure (say "prepare a release" or "check a page's SEO") can be saved as a skill or a slash command and run with one word next time.
It can run several agents at once. On bigger tasks Claude Code can split work between subagents that search the code or handle independent parts in parallel.
Is Claude Code for non-programmers?
Partly. Someone without a programming background can install it and have it build a simple tool, especially through the desktop app. The trouble starts when you need to judge whether the result is any good.
The Stack Overflow Developer Survey 2025 found that 84% of respondents use or plan to use AI tools in their development process, and 51% of professional developers use them daily. The same survey also showed trust in the accuracy of AI output going down. That matches my experience: the agent is fast, but just as confident when it is wrong. I described four such cases that each cost me a day in Four bugs where the tool lied.
So for non-programmers: yes for internal tools, prototypes and automations. For anything that handles other people's data or money, bring in someone who can read the code. The basic security minimum, even for prototypes, is in Security even when you're just prototyping.
Claude Code vs. Cursor and other AI tools
| Claude Code | Cursor | Lovable and similar | |
|---|---|---|---|
| Where it runs | terminal, desktop, VS Code, JetBrains, web | its own editor (a VS Code fork) | browser |
| Main strength | independent work across the whole project, shell, MCP | fast edits right in the editor | a first version of an app without code |
| Control over code | full, works in your repository | full | more limited, export to GitHub |
| Who it's for | developers and advanced users | developers | beginners, quick prototypes |
These tools are not mutually exclusive. I used to start in Lovable and finish in Cursor, as I described in How I build apps and websites using AI coders. Today I do most of the work, from the first line to deployment, in Claude Code, because it sees the whole project and the tools around it.
Five habits that decide the outcome
- Small, clearly bounded tasks. "Fix login" is a worse brief than "after logout the user lands on /dashboard instead of /login, fix it and add a test".
- Context in
CLAUDE.md, not in your head. Anything you explain to the agent a third time belongs in the file. - Verification, not trust. Have the agent run the tests or the app and show the result. "Done" without proof is not done.
- Few tools. Every extra MCP server is one more option the agent has to choose between. Connect only what the job needs.
- Permissions with care. Auto-approving commands is convenient, but keep it to places where a mistake breaks nothing.
Frequently asked questions
Is Claude Code free? No. The free Claude plan does not include it. You need Pro (20 USD a month, 17 USD billed annually), Max, Team, Enterprise, or API billing.
Does Claude Code work in languages other than English? Yes. You can write instructions in Czech or another language and it answers in that language. I still keep code, variable names and commit messages in English, because that is what most documentation and libraries use.
Do I need to know how to code? Not for simple tools and prototypes. For a production app that handles user data, you should understand what the agent changed, or have someone who can review it.
Does Claude Code run on Windows? Yes, natively on Windows 10 and 11 via PowerShell, and in WSL. Installing Git for Windows is recommended for full shell support.
What is the difference between Claude Code and Claude in the browser? Claude in the browser is a chat that only sees what you paste in. Claude Code works directly in your project: it reads and edits files, runs commands, and connects to other tools through MCP.
