Menu
Products

Claude Code

4 min readbeginnerUpdated 28 Sept 2026
1 · In one line

Claude Code is Anthropic's coding assistant that reads a project, edits files and runs commands for you, asking before risky steps.

1 · What it is

Claude Code is a coding tool from Anthropic. It first appeared in February 2025 as a limited research preview that ran in the command line, the text window where you type commands. You describe a job in everyday words. It can handle routine tasks, explain confusing code and take care of git, the tool programmers use to save versions of their work.

Inside, a Claude model does the thinking. Sonnet suits most coding work, and Opus is meant for harder design choices. Claude Code wraps that model. It hands the model tools, such as reading files and running commands, and decides what information the model sees.

Because it can act, it has a permission mode, a setting for what it may do without asking. In plan mode it only explores and suggests a plan, without editing your code. Stay alert to prompt injection, where someone slips in text that tries to trick an AI into following their instructions instead of yours.

2 · Why it exists

A chatbot that only writes text leaves three jobs to you.

Text is not actionWithout tools, a model can only reply with words; it cannot open your files or run your tests.
One file is too narrowA real bug fix often means reading several files, editing them together and running the tests again.
Actions carry riskA tool that can run any command you could run needs a way for you to stay in control.
3 · How it works

Follow one request, such as fix the failing tests.

How Claude Code works through one task A request and the project files with CLAUDE.md feed a loop: gather context, a permission check that asks you first, take action, verify results. At the permission step you approve or say no. If the task is not done, verify loops back to gather. When it is done, you review the changes. ONE TASK IN CLAUDE CODE Your request Gather context Permission Take action Verify You decide You review fix the failing tests search, read files and CLAUDE.md asks first unless your mode allows edit files or run a command run the tests again you approve or say no read the diff before you keep it not done yet done
  1. 1 · askYou describe the task in plain words in the terminal, an editor, the desktop app or the browser.
  2. 2 · gatherClaude Code searches and reads the relevant files, plus the project notes in CLAUDE.md.
  3. 3 · actIt edits files or runs commands, asking you first unless your permission mode allows the step.
  4. 4 · verifyIt runs the tests again and repeats the loop until the task is done or you step in.

The model does the reasoning; Claude Code is the layer of tools and permissions around it.

4 · Where it's used
WhoWhat they askWhat it works with
Student“Explain what this project does and where the main function lives.”The files in the project folder
Developer“Fix the failing test in the login module and show me what changed.”Test output, source files and the shell
Team lead“Follow the rules in CLAUDE.md and rename this function everywhere.”Project instructions and every file that uses the name
5 · What it solves, and what it doesn't
solves
  • It can change many files together, instead of seeing only one file.
  • It checks its own work by running commands such as tests.
  • It saves a copy of each file before editing, so you can undo a bad change.
  • It can reach outside tools and data through MCP servers, add-ons that connect it to other services.
doesn't solve
  • Those saved copies cannot undo changes on other computers, such as a database or a live website.
  • You are still responsible for reviewing code and commands before you approve them.
  • A new session does not remember the previous conversation on its own.
  • Notes in CLAUDE.md guide it but are not strictly enforced rules.
6 · Go deeper
AI agentsMCPAgentic workflowsPrompt injection

Sources used

This explainer is written in original language. The links below support its factual claims.

  1. docsClaude Code overview, Anthropic · read 28 Sept 2026
  2. docsHow Claude Code works, Anthropic · read 28 Sept 2026
  3. docsSecurity, Anthropic · read 28 Sept 2026
  4. docsHow Claude remembers your project, Anthropic · read 28 Sept 2026
  5. docsConnect Claude Code to tools via MCP, Anthropic · read 28 Sept 2026
  6. repoanthropics/claude-code, Anthropic · read 28 Sept 2026
  7. officialClaude 3.7 Sonnet and Claude Code, Anthropic · read 28 Sept 2026
Suggest a correction