Claude Code is Anthropic's coding agent: it reads your project, edits files and runs commands, from the terminal first, with extensions for VS Code and JetBrains IDEs. Getting it running takes a few minutes — install it, sign in with an account that includes it, give it a CLAUDE.md with your project's rules, and connect the tools you want it to use through MCP. Every step here follows Anthropic's Claude Code documentation as we read it on 2026-11-07; the official docs are the authority when anything differs.
1. Check the requirements
From the setup docs: macOS 13.0 or later; Windows 10 1809+ or Windows Server 2019+; Ubuntu 20.04+, Debian 10+ or Alpine Linux 3.19+; at least 4 GB of RAM; an x64 or ARM64 processor. On Windows, Claude Code runs natively or inside WSL — Git for Windows is optional (without it, it uses PowerShell), and sandboxing works only under WSL 2.
2. Install
The docs label the native installer "Recommended":
# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash
# Windows PowerShell
irm https://claude.ai/install.ps1 | iex
"Native installations automatically update in the background." The alternatives:
| Method | Command | Updates |
|---|---|---|
| Homebrew | brew install --cask claude-code |
Manual (brew upgrade) — "Homebrew installations do not auto-update" |
| WinGet | winget install Anthropic.ClaudeCode |
Through WinGet |
| npm | npm install -g @anthropic-ai/claude-code |
Through npm; needs Node.js 22 or later |
A note on npm, because sources disagree: the setup docs list it as supported with Node.js 22+, and the package was current on npm (version 2.1.319, published 2026-11-06), but the project's GitHub README calls npm installation deprecated and still shows an old Node.js 18 badge. Use the native installer unless you have a reason not to.
Then open a terminal in a project and run claude.
3. Sign in with the right account
Claude Code is not part of the free plan. The authentication docs: "Claude Code requires a Pro, Max, Ultra, Team, Enterprise, or Console account. The free claude.ai plan does not include Claude Code access." The ways to sign in:
- a Claude subscription (Pro, Max or Ultra) — log in with your claude.ai account;
- a Claude Console account, billed by API usage, or an
ANTHROPIC_API_KEYenvironment variable; - a cloud provider: Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry.
Prices on claude.com/pricing on 2026-11-07:
| Plan | Price | Claude Code |
|---|---|---|
| Free | $0 | No |
| Pro | $20 a month, or $17 a month billed yearly ($200) | Yes |
| Max | $100 a month | Yes, 5× Pro usage |
| Ultra | $200 a month | Yes, 10× Pro usage a week, no 5-hour limit |
| Team | per seat, for 2–150 people (we could not confirm the US-dollar seat price on that date) | Yes |
| Enterprise | $20 per seat a month (annual, 20-seat minimum) plus usage at API rates | Yes |
Subscription plans have usage limits that reset over time; a Console account pays per token instead. For occasional use, Pro is the entry point.
4. Give it a CLAUDE.md
The memory docs: "CLAUDE.md files are markdown files that give Claude persistent instructions… Claude reads them at the start of every session." The useful locations:
./CLAUDE.md(or./.claude/CLAUDE.md) — the project's rules, committed with the code;~/.claude/CLAUDE.md— your personal rules for every project;./CLAUDE.local.md— personal notes for one project, kept out of git.
Start short: how to build and test, the package manager, code style, folders not to touch, and "never commit secrets". Add a line whenever Claude repeats a mistake. Claude Code can also read an AGENTS.md — the file Cursor and other agents use — without any setup, from version 2.1.277, but only when there is no CLAUDE.md in the folder or above it.
5. Use it in your editor
- VS Code: the extension (
anthropic.claude-code, VS Code 1.94.0 or later) gives Claude Code "a native graphical interface… integrated directly into your IDE". The docs page also offers an install for Cursor. - JetBrains: a plugin for IntelliJ IDEA, PyCharm, WebStorm and others; "The plugin requires the Claude Code CLI, installed separately."
6. Connect your first MCP server
MCP servers give Claude Code tools beyond your files — an issue tracker, a database, a browser, an image service (what an MCP server is). The MCP docs use one command for a remote server:
claude mcp add --transport http <name> <url> --header "Authorization: Bearer <token>"
Scopes decide who gets the server: --scope local (the default, this project, stored in ~/.claude.json), --scope user (all your projects) and --scope project (a shared .mcp.json in the repository). SSE servers are deprecated in favour of HTTP. Inside a session, /mcp shows each server's status, signs in with OAuth and reconnects.
To give Claude Code our image tools — render_image, get_image, list_templates, get_template, create_signed_url, get_usage, generate_template, take_screenshot, auto_og_url, create_batch, get_batch — for rendering images from templates and taking screenshots of public pages:
export OG_API_KEY="sk_live_…"
claude mcp add --transport http shareglint https://shareglint.com/mcp \
--header "Authorization: Bearer $OG_API_KEY"
Then ask, for example: "Add an og:image to every blog post using the blog-og template." The Claude Code MCP guide covers keeping the key out of a shared .mcp.json and a full worked example on an Astro blog.
7. Habits that pay off
- Review diffs before you accept them, especially in code that handles money, auth or personal data.
- Commit before big tasks, so any wrong turn is one
git restoreaway. - Keep tool approvals for anything that spends or deletes, and allow only read-only tools to run unasked.
- Give the agent its own API keys for outside services, so you can revoke them without touching production.
Choosing between Claude Code and an AI editor? Our Claude Code vs Cursor comparison puts the plans, models and workflows side by side.
