What Is Claude Code?
A beginner-friendly guide to Anthropic's coding agent: what it is, how it differs from chatting with Claude, where you can use it, what it can do with your project, how permissions keep you in control, and how to start safely.
Imagine two ways of getting help with a leaky pipe. In the first, you phone a plumber, describe the problem and follow their spoken advice while you crouch under the sink yourself. In the second, the plumber comes to your house, looks at the pipes, tightens the joint, tests it and tells you what they did. Both involve the same expertise, but the second plumber can actually see your pipes and work on them.
That difference is a good way to understand Claude Code. Chatting with Claude is like the phone call: Claude advises, and you do the work. Claude Code is like the plumber in your house: it can look at your project, make changes and check the results, while you stay in charge. This guide explains it from scratch. No technical knowledge is needed. Every new word is explained when it first appears, and each big idea comes with an everyday comparison.
Claude Code changes very quickly, and many guides online become out of date within weeks. Plans, prices, supported tools, settings and features have all shifted during 2026. The details here reflect Anthropic’s documentation and recent coverage at the time of writing. Before you install or pay for anything, check the current Claude Code overview at code.claude.com and the pricing page at claude.com/pricing.
1What Is Claude Code?
Let us begin with the official description and then unpack it.
Anthropic’s documentation describes Claude Code as an agentic coding tool that reads your codebase, edits files, runs commands and integrates with your development tools. It is available in your terminal, in code editors, in a desktop app and in the browser.
Several words in that sentence deserve explanation. A codebase is the whole collection of files that make up a software project. Agentic means that the tool does not only answer questions but takes actions, working through a task step by step, much like an agent acting on your behalf. A command is a typed instruction that tells a computer to do something, such as running a program or testing your code. And a development tool is any program developers use to build software, such as version control systems or test runners.
Codebase
All the files that make up a software project, such as the code, settings and documentation.
Agentic
Able to take actions and work through a multi-step task, not just reply with text.
Terminal
A text window where you type commands to control your computer, also called the command line.
IDE
Integrated development environment: a code editor with built-in tools, such as VS Code or JetBrains editors.
A cookery book tells you how to make bread. A skilled baker in your kitchen can look at your oven, mix the dough and adjust the timing. Claude Code is the baker, not the book, because it works directly with your project.
Claude Code is the same Claude intelligence, given the ability to act on your files and tools, with your permission, instead of only chatting.
2Claude Code Versus Chatting with Claude
The two are related but built for different jobs, so it helps to see them side by side.
In an ordinary Claude chat, you describe a problem or paste in some code, and Claude replies with text. You then copy the answer to wherever you need it and test it yourself. This works well for questions, explanations and small snippets. In Claude Code, Claude has project-level access. It can read the relevant files itself, make changes across several files, run your tests, see the results and try again if something fails.
| Aspect | Chat with Claude | Claude Code |
|---|---|---|
| What Claude sees | Only what you paste or upload | Your project files, when you let it |
| What Claude does | Replies with text and files in the chat | Edits files and runs commands, with permission |
| Who tests the result | You, by copying and trying it | Claude can run tests and react to the output |
| Best for | Questions, explanations, drafts, small snippets | Changes that touch real projects across several files |
| Your role | Copy, paste and check | Direct, review and approve |
flowchart LR
A["Chat: you paste code"] --> B["Claude replies with advice"]
B --> C["You copy, apply and test it"]
D["Claude Code: you describe the goal"] --> E["Claude reads files and plans"]
E --> F["Edits files and runs checks"]
F --> G["You review the changes"]
Asking a friend for directions over the phone is chat. Having the friend sit in your passenger seat, watch the road and tell you where to turn is Claude Code. The friend sees what you see, so the help is more precise.
Guides describe people such as designers and analysts also using Claude Code, for example to turn specifications into code or to process and summarise data. The terminal version has a learning curve, while the desktop and web versions are described as more accessible entry points.
3Where You Can Use It
Claude Code runs in several places, and they share the same underlying engine.
Anthropic’s documentation says that Claude Code runs on several surfaces: the terminal, IDE extensions, a desktop app and the web. The terminal command-line tool, often called the CLI (short for command-line interface), is the original and fullest experience. Extensions exist for editors such as VS Code, and a plugin exists for JetBrains editors. The Claude desktop app has a Code tab, and there is also a web version at claude.ai/code, which is described as available on desktop browsers and in the Claude app for iOS and Android.
Terminal
The full command-line tool for working directly in your terminal. Often where developers begin.
Code editors
Extensions for VS Code and a plugin for JetBrains editors, so Claude sits beside your code.
Desktop app
A graphical app where you can review changes visually and run several sessions side by side. A paid subscription is required.
Web and mobile
Start tasks in the browser or the Claude mobile app, and in some cases move them to your terminal later.
flowchart TD
A["Claude Code engine"] --> B["Terminal"]
A --> C["VS Code and JetBrains"]
A --> D["Desktop app: Code tab"]
A --> E["Web and mobile app"]
A --> F["Automation: CI/CD, chat and browser workflows"]
A good coffee can be served at the counter, in a takeaway cup or delivered to your desk. The coffee is the same, and you choose the way that fits your day. Claude Code works the same way across its surfaces.
Anthropic’s documentation describes ways to hand work between them, such as starting a long task on the web and pulling it into your terminal later. Features like this change, so check the current overview.
4What Claude Code Can Do
The range of tasks is wide, because working on a project involves many small activities.
Understanding an unfamiliar project
Ask Claude to explain how a project is organised, where a feature lives or what a confusing section does. It can read the files rather than guess.
Building features
Describe what you want in plain language, and Claude can plan the change, edit the files involved and check that it works.
Fixing bugs
Share an error message or a description of the problem. Claude can investigate the code, find likely causes, propose a fix and run tests to confirm.
Running tests and fixing failures
Claude can run your test suite, read the failures and try again, looping until the checks pass or it needs your help.
Tidying and improving code
Rename things consistently, restructure messy sections or update older code, across several files at once.
Working with version control
Documentation describes Claude creating commits and pull requests, which are ways of recording and proposing changes to a project.
Automating chores
The documentation describes using Claude Code to automate work you keep putting off, including scheduled tasks that run on Anthropic-managed infrastructure.
A capable site assistant can read the building plans, fix a fault, test the result and file the paperwork. They do not do only one trick; they handle the many small tasks around a job. Claude Code does the same around a software project.
Claude Code can make mistakes, just like any assistant. Review changes before accepting them, especially for anything important.
5How It Works: Plan, Act, Check
A typical session follows a simple rhythm that you will start to recognise.
You describe the goal
You write what you want in ordinary language, such as fixing a bug or adding a small feature.
Claude explores
It searches and reads the relevant files to understand how the project fits together.
Claude proposes a plan or acts
It works out the steps and begins editing files or running commands, asking for approval as set by your permissions.
Claude checks its work
It may run tests or other checks, read the output and adjust.
You review and decide
You look at the changes, ask for adjustments or accept them.
flowchart TD
A["Your goal"] --> B["Explore the project"]
B --> C["Plan the steps"]
C --> D["Make changes"]
D --> E["Run checks"]
E --> F{"Did it work?"}
F -- "No" --> C
F -- "Yes" --> G["You review and accept"]
A careful carpenter looks over the job, plans the cuts, builds, then measures to check that the result is square. If it is not, they adjust and measure again. Claude Code follows a similar loop of planning, acting and checking.
For bigger changes, ask Claude to propose a plan first and agree it with you before it edits anything. A short conversation about the plan often prevents wasted work.
6Staying in Control: Permissions
Because Claude Code can change files and run commands, it is designed to ask before it acts.
A permission is your approval for Claude to do something specific. In Claude Code, actions that could change your files or run commands are generally subject to approval, according to how you configure it. You can approve actions one at a time, allow certain safe actions in advance, or deny others entirely. The details of how permission modes work have varied across versions, so it is best to read the current settings documentation and start with the more cautious options.
Ask before acting
Claude pauses and asks you to approve file changes or commands before they happen.
Allow safe actions
You can allow specific low-risk actions in advance so you are not asked each time.
Deny risky actions
You can block certain commands or files, such as those holding secrets, from being touched.
Review the changes
You can inspect what changed before accepting it, including visual comparisons in some versions.
Why permissions help
- You stay in charge of what happens to your files
- Risky actions can be blocked in advance
- It builds trust gradually, step by step
- It limits the harm if Claude misunderstands a request
Trade-offs to know
- Frequent approval prompts can slow you down
- Allowing too much too soon removes the safety net
- Settings can be confusing at first
- Defaults and modes differ across versions
A new contractor might need to ask before touching anything in your house. As trust grows, you let them handle routine jobs on their own but still check anything that could be costly or hard to undo. Permissions let you set that level of trust.
7Teaching Claude Code About Your Project
Claude Code starts each session without remembering the last one, so projects can provide standing notes.
Guides describe a special file called CLAUDE.md that Claude Code reads automatically at the start of a session. It acts as a standing set of instructions for a project: how to run the tests, how the project is organised, what coding style to follow and what to avoid. Because Claude reads it every time, you do not need to repeat those details in every session.
Claude Code also has other ways to extend and customise it. The documentation and coverage mention hooks, which run your own checks automatically at certain moments, MCP servers, which connect Claude to outside tools and data, subagents, which are helper agents that handle parts of a task, skills, which are reusable playbooks, and plugins, which bundle such extras. A newcomer does not need any of these to begin.
| Extra | What it does | Beginner need? |
|---|---|---|
| CLAUDE.md | Standing project notes read at the start of each session | Very useful early on |
| Hooks | Run your own checks or actions automatically at set moments | Later |
| MCP servers | Connect Claude to outside tools and data | Later |
| Subagents | Helper agents that take on parts of a larger task | Later |
| Skills and plugins | Reusable playbooks and bundles of extras | Later |
When a new team member joins a project, a good lead gives them a one-page guide: how things are laid out, how to run the checks and what the house rules are. CLAUDE.md is that one-page guide for Claude.
A tidy, focused set of project notes works better than a long, rambling one. Include what Claude would otherwise keep asking about, and update it as the project changes.
8Getting Started
The first steps are simple, and Anthropic’s quickstart guides cover each environment.
Most Claude Code surfaces need a Claude subscription or an Anthropic Console account, according to the overview, and the terminal tool and VS Code also support some third-party providers. For the desktop app, a paid subscription is required. Because exact installation steps differ by system and change often, follow the current instructions at code.claude.com rather than copying commands from older articles.
Choose your surface
Pick the terminal, a code editor, the desktop app or the web, depending on what feels comfortable.
Check your plan or account
Make sure you have a suitable subscription or Console account for the surface you picked.
Install and sign in
Follow the official quickstart for your system, then sign in with your Claude account.
Start with a safe practice project
Try it on a small or copied project first, with version control in place so changes can be undone.
Begin with questions
Ask Claude to explain the project before asking it to change anything, to see how it reads your files.
Make small changes and review
Try a small, clear task, read the changes carefully and build up from there.
When you learn to drive, you start in an empty car park before taking the motorway. A small practice project is your empty car park for Claude Code.
Before letting any tool change your files, make sure you can undo it, for example with version control or a copy of the project.
9Writing Good Requests
Clear requests lead to better, faster results, just as with any assistant.
State the goal
Say what you want to achieve and why, in plain language.
Point to the right place
Mention the file, feature or error so Claude starts in the right area.
Say what success looks like
For example, which tests should pass or what behaviour you expect.
Set limits
Say what must not change, and ask Claude to check with you before big decisions.
Ask for a plan first
For larger tasks, have Claude outline its approach before editing.
Work in small steps
Break big jobs into smaller tasks that are easier to review.
Vague request
“Fix the login.”
Clear request
“Users sometimes see a blank page after signing in with a saved password. Find the cause, propose a fix and show me the plan before you change anything. Do not touch the payment code. When done, run the login tests and tell me the result.”
Telling a mechanic “make it better” gets guesswork. Telling them “there is a rattle from the front left wheel at low speed” gets a diagnosis. Specific symptoms lead to specific fixes.
10Plans, Usage and Costs
Claude Code is not a separate purchase for most people, but it does use up allowance.
According to Anthropic’s documentation and coverage, Claude Code is included in paid Claude plans, such as Pro and Max, and can also be used with an Anthropic Console account, where you pay by usage. Coverage notes that on subscription plans, usage is generally shared across Claude chat, the desktop app and Claude Code, rather than having a separate allowance. Heavier use, such as long sessions or several helper agents running at once, consumes more.
Prices, usage limits and plan features have changed more than once in 2026, and some coverage notes changes in how certain non-interactive uses are counted. Because of this, avoid relying on any price quoted in an article without checking the date, and confirm current numbers on Anthropic’s pricing page.
| Way to pay | How it works | Good to know |
|---|---|---|
| Subscription plan | A fixed monthly fee with usage limits, shared across Claude products | Heavy use can hit limits and need a higher plan |
| Console account | Pay by usage through Anthropic’s developer platform | Costs scale with how much you use, so monitor them |
| Team and Enterprise | Seat-based plans with administrative controls | Features and seat types vary, so check with your administrators |
A mobile plan with a monthly data allowance covers ordinary use, but streaming films all day uses it faster. Light Claude Code use fits easily, while long, intensive sessions use more of your allowance.
Claude Code includes ways to check how much you have used. Keep an eye on it, particularly when trying long or parallel tasks for the first time.
11Safety and Privacy
A tool that can read files and run commands deserves some sensible precautions.
Protect secrets
Keep passwords, keys and private files out of reach, for example by blocking Claude from reading files that contain them.
Use version control
Keep your project under version control so any change can be reviewed and undone.
Trust sources carefully
Be cautious with outside tools, plugins and skills, since they can influence what Claude does.
Follow work policies
If your code belongs to an employer, follow its rules about AI tools and what code may be shared.
One risk worth understanding is prompt injection. This happens when text inside a file, a web page or a message tries to give Claude hidden instructions. Because Claude reads many kinds of content while working, it is wise to review what it plans to do, particularly if it has processed material from untrusted sources, and to keep its permissions tight.
flowchart TD
A["Untrusted text in a file or web page"] --> B["May contain hidden instructions"]
B --> C{"Do permissions and review stand in the way?"}
C -- "Yes" --> D["Risky action is blocked or noticed"]
C -- "No" --> E["Risk of unintended action"]
A conscientious assistant who follows written notes might be tricked by a forged note slipped into a pile of papers. Checking unusual instructions, and keeping the keys to the vault locked, protects you.
12Who Uses Claude Code and For What
Although it began with developers, its uses are broadening.
Professional developers
Use it to understand large codebases, build features, fix bugs, write tests and handle routine chores faster.
Learners
Use it to understand how existing code works, with Claude explaining files and ideas in plain language.
Designers and product people
Guides describe turning specifications into working prototypes with Claude’s help.
Analysts
Guides describe using it to process and summarise data and to automate repeated tasks.
Teams
Teams use it within shared workflows, including connections to chat tools and automation pipelines, depending on how administrators set it up.
A good workshop tool is bought by professional builders but also helps hobbyists and students. Claude Code serves a similar spread of people, with the level of care depending on what the project is worth.
The more important or sensitive the project, the more careful your review should be. Never let a tool push changes into critical systems without a human check.
13Common Mistakes
These errors appear often, and each has an easy fix.
Problem
Granting broad permissions on day one to avoid approval prompts.
Why It’s Harmful
It removes the safety net before you understand how Claude behaves, so a misunderstanding can cause real damage.
Correct Approach
Start cautiously, approve actions as they come up and allow only routine, low-risk actions in advance.
Problem
Working on a project with no version control or backup.
Why It’s Harmful
If a change goes wrong, there may be no easy way to see what happened or to undo it.
Correct Approach
Keep your project under version control and commit before large changes.
Problem
Giving huge, vague tasks in a single request.
Why It’s Harmful
Large vague requests lead to sprawling changes that are hard to review and easy to get wrong.
Correct Approach
Break work into small, clear steps, and ask for a plan before editing.
Problem
Accepting changes without reading them.
Why It’s Harmful
Mistakes slip into your project, and you may not understand your own code afterwards.
Correct Approach
Review each change, run the checks yourself and ask Claude to explain anything unclear.
Problem
Following an old guide with outdated commands, prices or settings.
Why It’s Harmful
Claude Code changes quickly, so older steps may fail or mislead you about cost and features.
Correct Approach
Check the date on every guide and rely on the current documentation at code.claude.com.
Problem
Leaving secrets within reach of Claude, such as keys in project files.
Why It’s Harmful
Secrets can be read, shown or mishandled, and may be exposed in outputs or logs.
Correct Approach
Keep secrets out of project files and block access to files that contain them.
14Benefits and Trade-offs
Here is the whole picture in one place.
Advantages
- Works directly with your real project files
- Can plan, edit, run checks and retry
- Available in terminal, editors, desktop app and web
- Permissions keep you in control of actions
- Extensible with project notes, hooks, connectors and skills
Disadvantages / Trade-offs
- Output must be reviewed, since mistakes can occur
- Usage and costs can add up with heavy use
- The terminal version has a learning curve
- Security needs care with secrets and untrusted content
- Features, plans and guides change quickly
A power tool speeds up the work, but a careful craftsperson still measures, checks and takes responsibility for the finished piece. The same applies when you work with Claude Code.
15Frequently Asked Questions
It is Anthropic’s agentic coding tool that reads your project, edits files, runs commands and works with your development tools, available in the terminal, editors, a desktop app and on the web.
In chat, Claude sees only what you paste and replies with text. In Claude Code, Claude can read your project files, make changes and run checks, with your permission.
Not necessarily. The terminal version has a learning curve, but the desktop and web versions are described as more accessible. Some non-developers use it for tasks like prototyping or data work. You should still review what it does.
In the terminal, in editors such as VS Code and JetBrains editors, in the Claude desktop app’s Code tab, and on the web at claude.ai/code, including through the Claude mobile app.
Coverage says Claude Code is included in paid plans such as Pro and Max, with usage generally shared across Claude products, and that it can also be used through a Console account billed by usage. Plans and limits change, so check claude.com/pricing.
It is designed to ask for approval for actions that change files or run commands, depending on how you configure permissions. Start with cautious settings and review changes before accepting them.
A file of standing notes about your project that Claude Code reads at the start of each session, such as how to run tests, how the project is organised and what conventions to follow.
It can be, with care: use version control, protect secrets, keep permissions tight, review changes and follow your organisation’s policies about AI tools and code sharing.
They are optional extras. Hooks run your own checks automatically, MCP servers connect Claude to outside tools and data, subagents handle parts of a task, and skills are reusable playbooks. Beginners can ignore them at first.
The Claude Code overview and quickstarts at code.claude.com, and the pricing page at claude.com/pricing. This tool changes quickly, so check the date on any guide.
16Summary and Key Takeaways
Claude Code is Anthropic’s agentic coding tool: the same Claude intelligence, given the ability to read your project, edit files, run commands and work with your development tools, with your approval. Unlike a chat, where you copy and test code yourself, Claude Code works inside your project through a loop of exploring, planning, acting and checking, and you review the result. You can use it in the terminal, in code editors, in the desktop app’s Code tab and on the web, and it grows with optional extras such as project notes, hooks, connectors, subagents and skills. Permissions keep you in control, and good habits such as version control, small steps, clear requests, protected secrets and careful review keep it safe. It is included in paid Claude plans or available through a Console account, though prices, limits and features change quickly, so rely on current documentation. Used thoughtfully, it takes on the legwork of building software while you keep the judgement.
Key Takeaways
- It acts on your project — Claude Code reads files, edits them and runs commands, unlike a chat that only replies with text.
- Several ways in — terminal, code editors, the desktop app’s Code tab and the web, all built on the same engine.
- It works in a loop — explore, plan, act and check, with you reviewing the outcome.
- Permissions protect you — start cautious, approve actions, allow only low-risk ones in advance and block sensitive files.
- Project notes help — a CLAUDE.md file gives Claude standing instructions, while hooks, connectors, subagents and skills can wait until later.
- Start small and safe — use version control, practise on a small project, ask questions first and make small changes.
- Review and verify — mistakes can happen, costs can add up and guides go stale, so check results and rely on current official documentation.