Claude Code Complete Tutorial: From First Prompt to Agents, MCP, Hooks and Automation

What Claude Code is

Claude Code is an agent that works inside your project: it reads files, runs commands, edits code, runs tests and opens pull requests, while you watch and approve. It runs in the terminal, in VS Code / JetBrains, in the Claude desktop app, on the web (cloud sessions connected to GitHub) and on mobile, and it is used by non-engineers too: PMs build prototypes and analyses, analysts clean data and build reports.

Install and first session

Context: CLAUDE.md and how to brief Claude

CLAUDE.md files are loaded automatically: the project one (shared in git), personal ones (~/.claude/CLAUDE.md) and folder-level ones. Put in them: how to build and test, code style, architecture notes, "never do" rules, and how to publish. Keep them short and specific.

Good prompts give the goal, constraints and how to verify:

Goal: add a "download CSV" button to the orders page.
Constraints: reuse the existing Table component; no new dependencies; Indian date format.
Done when: the button downloads the filtered rows, a unit test covers the CSV builder, and `npm test` passes.
Plan first, then implement.

Use @path/to/file to point at files, paste screenshots for UI bugs, and ask Claude to read docs URLs.

Plan mode, permissions and checkpoints

Built-in slash commands worth knowing

Extending Claude Code

Custom slash commands and Skills

• Custom commands: Markdown files in .claude/commands/ (project) or ~/.claude/commands/ (personal); /release-notes can run your team's exact release-notes prompt with $ARGUMENTS.
• Skills: folders with a SKILL.md (instructions + optional scripts) that Claude loads automatically when relevant, e.g. "our PRD format", "brand slide style", "how to deploy". Plugins bundle commands, skills, agents, hooks and MCP servers for sharing.

MCP servers

claude mcp add connects tools (GitHub, Jira, Sentry, Postgres, Figma, Playwright browser, your own). Claude can then read a Figma frame and build the UI, query a read-only database to answer a data question, or open a browser to test the app.

Hooks

Hooks run your own commands at moments in Claude's workflow (before or after a tool call, when a session ends). Examples: run the formatter after every edit, block edits to .env files, send a Slack message when a long task finishes.

Subagents

Specialised agents with their own instructions, tools and context window, e.g. a "code reviewer", "test writer" or "data analyst". Claude delegates to them, keeping the main conversation focused.

Automation

Best practices

practices = {"Explore → plan → code → verify": "Ask Claude to read and plan before editing; finish with tests or a demo",
             "Give it a way to check its work": "Tests, a build command, screenshots, a sample input/output",
             "Small, reviewable steps": "One feature per session; commit often; use branches",
             "Keep context clean": "/clear between unrelated tasks; CLAUDE.md for durable rules",
             "Never paste secrets": "Use env vars and secret stores; review diffs before pushing"}
for k, v in practices.items(): print(f"- {k}: {v}")

- Explore → plan → code → verify: Ask Claude to read and plan before editing; finish with tests or a demo
- Give it a way to check its work: Tests, a build command, screenshots, a sample input/output
- Small, reviewable steps: One feature per session; commit often; use branches
- Keep context clean: /clear between unrelated tasks; CLAUDE.md for durable rules
- Never paste secrets: Use env vars and secret stores; review diffs before pushing

For PMs and analysts

• Prototype: "Build a clickable prototype of the onboarding flow in this Figma file using our component library," then share a preview link.
• Data analysis: point Claude at CSVs or a read-only database and ask for cleaned data, charts and a written summary.
• Specs and research: read the codebase to answer "how does pricing work today?" before writing a PRD.
• Automation scripts: small scripts that pull data from APIs into Sheets or Slack.

Summary

Learn more

• Claude Code documentation
• Claude Code best practices (Anthropic engineering)
• Anthropic Academy courses