How to Install Claude Code: A UK Business Guide (2026)

Claude Code installs in under five minutes on macOS, Linux, or Windows. You need Node.js 22 LTS, an Anthropic account on the Pro plan or higher, and a terminal you actually use. Run npm install -g @anthropic-ai/claude-code, then claude to authenticate through your browser. The first task you give it should not be code — it should be writing your project's CLAUDE.md. That file determines whether your team gets ten percent or sixty percent of the productivity gains, and most rollouts skip it.

Last updated: May 2026 · Covers Claude Code v2 (PowerShell-native on Windows) · Opus 4.7, Sonnet 4.6, Haiku 4.5

• Install command: npm install -g @anthropic-ai/claude-code (one line, every platform)
• Requires Node.js 22 LTS, Anthropic Pro plan (£17/mo), and a terminal
• Authenticate by running claude once and approving in your browser — no API key copying required
• Windows users no longer need WSL: Claude Code v2 ships native PowerShell support
• The five minutes you save by skipping the CLAUDE.md will cost you forty hours of bad output later

Contents

• What you actually need before installing Claude Code
• How to install Claude Code on macOS
• How to install Claude Code on Linux
• How to install Claude Code on Windows
• First-run authentication
• The single most important step nobody talks about: write your `CLAUDE.md`
• Project
• Rules
• Stack conventions
• Forbidden patterns
• Deployment
• Verifying the install with a real task
• Common installation problems and how to fix them
• What to install next: MCP servers, skills, and hooks
• How Ampliflow installs Claude Code for UK clients

What you actually need before installing Claude Code

Five things, none of them surprising. Most failed installs fail at this checklist, not the install itself.

`text PREREQUISITES INSTALL FIRST RUN ------------- ------- --------- Node 22 LTS --+ Anthropic Pro plan | A real terminal +---> npm install -g ---> claude A Git repo to test | @anthropic-ai/ | 5 minutes --+ claude-code | v OAuth browser flow

v Write your CLAUDE.md (the file that determines whether your team gets 10% or 60% of the productivity gains)

v +---------+---------+ v v Read-only task Approved write (verify install) task (build trust) `

That's the floor. There is no special hardware requirement, no GPU, no Docker. Claude Code does its work by talking to Anthropic's models over HTTPS — your machine just runs the orchestrator.

One thing the docs don't emphasise: the quality of your terminal matters more than the speed of your machine. You're going to spend a lot of time looking at Claude Code's diffs and plans. Use a terminal with proper colour rendering (iTerm2, Windows Terminal, Alacritty, Ghostty), a monospace font you can read at small sizes, and a window large enough to see a forty-line diff without scrolling. The terminal is where the work happens.

How to install Claude Code on macOS

One command, then authenticate.

`bash npm install -g @anthropic-ai/claude-code `

If npm complains about permissions, you're using a system-installed Node. Fix that first:

`bash

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash source ~/.zshrc

nvm install 22 nvm use 22 nvm alias default 22

npm install -g @anthropic-ai/claude-code `

Verify the install:

`bash claude --version `

You should see claude-code/2.x.x. If you do, you're ready to authenticate.

How to install Claude Code on Linux

Identical to macOS for any modern distribution. Ubuntu 22.04+, Debian 12+, Fedora 39+, Arch — all work. The only Linux-specific gotcha is that some distributions ship Node 18 or 20 in their package manager, which is below the recommended LTS. Use nvm or fnm rather than apt install nodejs.

`bash

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash source ~/.bashrc

nvm install 22 nvm use 22

npm install -g @anthropic-ai/claude-code claude --version `

If you're running Linux on a server (no GUI), the browser-based authentication flow still works — Claude Code prints a URL, you copy it to a browser on any machine, and paste the resulting code back. We use this flow daily for our Hermes Agent deployments on Oracle Cloud.

How to install Claude Code on Windows

As of Claude Code v2 (March 2026), Windows runs natively in PowerShell — WSL is no longer required. This is a significant change. If you remember Windows being a second-class citizen, that's outdated.

PowerShell install (recommended)

Open PowerShell 7+ (download from github.com/PowerShell/PowerShell if you only have Windows PowerShell 5.1):

`powershell

winget install Schniz.fnm

Add-Content $PROFILE 'fnm env --use-on-cd | Out-String | Invoke-Expression' . $PROFILE

fnm install 22 fnm use 22 fnm default 22

npm install -g @anthropic-ai/claude-code claude --version `

WSL install (still supported)

If you're already a WSL user, the install works identically to the Linux instructions above. The choice between PowerShell-native and WSL is mostly about where the rest of your toolchain lives. If you do all your development inside WSL Ubuntu, stay there. If your dev tools are a mix of Windows-native (Cursor, VS Code with Windows extensions) and you only WSL for Docker, install Claude Code natively in PowerShell.

Windows-specific gotchas to know about

• Path separators in `CLAUDE.md`: use forward slashes (src/components/) not backslashes — Claude Code normalises both, but cross-platform consistency saves you when an article like this one shares a CLAUDE.md between Windows and macOS team members
• Long path names: Windows still defaults to a 260-character path limit. Enable long paths in Group Policy or via registry if your project has deep nesting
• Antivirus interference: Microsoft Defender occasionally flags Claude Code's process spawning as suspicious. Add the claude executable and your project directory to exclusions if you see slowness

First-run authentication

Run `claude` once. A browser opens. Approve. Done.

`bash claude `

The first invocation in any directory triggers the auth flow. Claude Code opens your default browser to a sign-in page on claude.com, you sign in with the same account as your Pro/Max/Team subscription, you click approve, and the CLI completes setup automatically. There is no API key to copy, no .env file to edit, no token to paste. The OAuth flow stores credentials in your OS keychain (macOS Keychain, Windows Credential Manager, Linux Secret Service).

If you're on a headless Linux server with no browser, Claude Code prints a URL and a one-time code. Open the URL on any other machine, paste the code, and the server-side claude process completes setup over a polling channel.

If you have multiple Anthropic accounts (a personal Pro plan and a work Team plan, for example), claude will prompt for which to use. You can switch accounts later with claude logout and claude login.

The single most important step nobody talks about: write your CLAUDE.md

The teams that see ten percent of the gains treat Claude Code as autocomplete. The teams that see fifty to sixty percent treat it as a new team member who needs a proper onboarding. The CLAUDE.md file is that onboarding.

CLAUDE.md is a plain Markdown file in the root of your project. Claude Code reads it on every task. It tells the agent what your project is, what conventions to follow, which commands are forbidden, and where the important context lives. Most of the gap between bad and great Claude Code usage is the quality of this single file.

Here is a minimal CLAUDE.md for a UK B2B SaaS:

`markdown

Project

Next.js 16 App Router + Convex + Stripe. UK B2B SaaS targeting estate agents. Dev server: PORT=3001 npm run dev (port 3000 is occupied by another project)

Rules

• British English throughout (colour, organisation, behaviour)
• Never run taskkill /IM node.exe (kills other dev processes)
• Never commit .env files or API keys to git
• Always run npm test before claiming a task is done
• Prefer editing existing files over creating new ones
• Path alias: @/ maps to src/

Stack conventions

• UI components: src/components/ui/ (shadcn) — never reinvent shadcn primitives
• API routes: src/app/api/
• Tests: Vitest + Playwright. NEVER mock the database — we test against a local Postgres
• Tailwind v4 (no JIT config file — config is in @theme directive in globals.css)

Forbidden patterns

• No console.log in production code (use the logger in src/lib/log.ts)
• No any in TypeScript — use unknown and narrow
• No client components fetching data — use server components or route handlers

Deployment

• Push to main triggers Vercel auto-deploy
• Maximum three pushes per session (Vercel build-minute budget)

`

Roughly thirty lines of plain English. No special syntax. Claude Code reads this on every task and follows it like a brief.

"The instructions you give the agent — CLAUDE.md, skill files, context files — are more valuable than the code the agent writes."

— Boris Cherny, Engineering Lead, Claude Code

The pattern Anthropic call "every mistake becomes a constraint" is the most important habit to build in your first fortnight. When Claude Code does something wrong, you do not just fix it — you also add the rule to CLAUDE.md. The context file grows. The mistakes stop repeating. By week three, your CLAUDE.md is doing the work of three pair-programming sessions per developer.

What goes in a good CLAUDE.md:

• Project description in two sentences (stack, audience, purpose)
• Forbidden commands (anything that hurts shared infrastructure or kills processes — we have learned this one the hard way after taskkill /IM node.exe killed two parallel Claude Code sessions)
• Path conventions (aliases, where things live)
• Testing rules (what to mock, what not to mock, how to run tests)
• Deployment rules (push limits, build-minute budgets, where things deploy)
• Coding style (anything your linter does not catch but reviewers always raise)

What does not go in a CLAUDE.md:

• General programming knowledge ("React components should be functional") — Claude Code knows
• Things your linter already enforces — let the tooling do its job
• Wishful thinking ("write beautiful code") — be specific or do not bother
• Secrets — never put credentials in CLAUDE.md

Skip this step and you will spend the next month wondering why other teams claim transformative results from a tool that, in your hands, mostly produces mediocre suggestions. The tool is not the variable. The brief is.

Verifying the install with a real task

Start with a read-only task. Build trust before letting Claude Code write.

In your project directory:

`bash claude

Show me the structure of this project. List the top-level directories,

the entry point, the test runner, and the build command. Do not modify anything. `

Claude Code should respond within a few seconds with a clear summary. If it does, the install is healthy and authentication is working. If you see an authentication error, run claude logout then claude login and retry.

Next, ask it to explain something:

`bash

Read src/lib/auth.ts and explain in plain English what this module does.

Identify any patterns I should be aware of when adding new auth flows. `

This tests whole-file reading and reasoning. The response should reference real names from your file — function names, exports, imports — not generic descriptions.

Finally, ask it to do something small:

`bash

Add a comment block at the top of src/lib/auth.ts explaining what the

module does, in three sentences. Show me the diff. Do not commit. `

This tests the write-and-show-diff cycle. Claude Code should propose an edit and wait for your approval before applying. This is the loop you will spend most of your time in.

If all three work, your install is production-ready. The remaining work is entirely about the CLAUDE.md and the patterns you build into it.

Common installation problems and how to fix them

Most installs fail in one of six ways. None of them are mysterious.

"command not found: claude"

The claude binary is not on your PATH. Either npm installed globally to a directory that is not on PATH, or you installed inside a Node version manager that has not been activated in your current shell.

Fix: `bash

npm bin -g

echo $PATH | tr ':' '\n' | grep -i npm

`

"EACCES: permission denied" on macOS or Linux

You are using a system-installed Node and trying to write to a system-protected directory. The fix is not sudo npm install — it is to use a Node version manager that installs into your home directory.

Fix: `bash

brew uninstall node # macOS sudo apt remove nodejs # Ubuntu/Debian

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash

source ~/.zshrc nvm install 22 nvm use 22 npm install -g @anthropic-ai/claude-code `

Node version mismatch

Claude Code v2 requires Node 18+, but recommends Node 22 LTS. If you are pinned to Node 16 by another project, use a version manager to switch contexts.

`bash

nvm use 22

fnm use 22 `

Both nvm and fnm support .nvmrc / .node-version files — drop one in your project root and the right Node version activates automatically when you cd into it.

WSL path errors on Windows

If you are running Claude Code inside WSL but pointing it at a Windows-mounted directory (/mnt/c/Users/...), you may see slow file reads or path normalisation issues. The fix is to clone your project into the WSL filesystem (~/projects/) and work entirely from there. WSL's filesystem performance on /mnt/c/ is materially slower than its native ext4.

API authentication errors after auth succeeds

Usually means your Anthropic plan does not include Claude Code access. Free Anthropic accounts cannot use Claude Code. Verify in console.anthropic.com that your subscription is active and includes Claude Code.

Corporate proxy or firewall blocking the install

If your organisation routes traffic through a corporate proxy, npm needs to know about it:

`bash npm config set proxy http://your-proxy:port npm config set https-proxy http://your-proxy:port `

Claude Code itself respects the standard HTTP_PROXY and HTTPS_PROXY environment variables. If your proxy injects its own root certificate, you may need to set NODE_EXTRA_CA_CERTS to point at the certificate file.

For UK businesses in regulated sectors (finance, legal, healthcare) running through Zscaler or similar gateways: the install works, but you will need to involve IT to whitelist the install domains and the runtime endpoints. Anthropic publish the full domain list at claude.com/network — share that with your network team.

What to install next: MCP servers, skills, and hooks

A bare Claude Code install is the engine. Skills, MCP servers, and hooks are the rest of the car.

MCP servers worth installing first

Model Context Protocol (MCP) servers are how Claude Code connects to external systems — your databases, your monitoring, your internal APIs, your file system, GitHub. The right three or four MCPs change Claude Code from a coding assistant into an operating environment.

The ones we install on day one for every UK client:

• `@modelcontextprotocol/server-filesystem` — explicit filesystem access with sandbox controls. Built-in for Claude Code v2 but worth knowing about
• `@modelcontextprotocol/server-github` — read issues, pull requests, file PRs without leaving the terminal. Needs a GitHub PAT
• `@playwright/mcp` — browser automation for verifying UI changes against a running dev server. Replaces a surprising amount of manual QA work
• `@modelcontextprotocol/server-postgres` — direct, read-only database queries from inside Claude Code. Works with any Postgres connection string

Install via your project's .mcp.json:

`json { "mcpServers": { "github": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-github"], "env": { "GITHUB_TOKEN": "ghp_..." } }, "playwright": { "command": "npx", "args": ["-y", "@playwright/mcp"] } } } `

For the broader 2026 stack of AI tools UK SMEs are deploying alongside Claude Code, see our AI tools stack for UK SMEs guide — Claude Code sits in the engineering layer; Hermes Agent in the operational layer.

Skills — your first one should be tiny

Skills are reusable prompts that encapsulate a domain — your code review checklist, your incident response playbook, your release process. Anthropic ship Claude Code with a skill loader; the community ships hundreds of skills you can drop in.

Your first skill should be small and specific. Not "write good code" — that is too vague. Try "follow our PR description format" or "run our specific deploy command sequence." A small skill that you actually use teaches you the system. A grandiose skill that nobody invokes wastes the slot.

The deep guide is in our Claude Code skills piece — it covers writing your own, sharing across a team, and the governance patterns that prevent your skill library becoming unmaintainable.

Hooks for safety guardrails

Hooks run before or after specific Claude Code actions. The two hooks every team should configure:

• PreToolUse hook that blocks dangerous shell commands (rm -rf, force-push to main, anything matching your forbidden-commands pattern)
• Stop hook that runs your test suite or linter before Claude Code claims completion

Hooks live in .claude/settings.json at the project root. They are bash one-liners or short scripts. The investment is twenty minutes; the upside is that your most catastrophic failure modes become impossible.

How Ampliflow installs Claude Code for UK clients

A three-stage rollout, every time. We have learned that compressing this timeline costs more time than it saves.

Stage 1 — Foundation week

A single engineer installs Claude Code on their machine, writes the first version of CLAUDE.md, and runs Claude Code against one low-risk track for a week. The track might be the test suite, documentation, or a specific refactor that has been on the backlog for months.

This week is for learning, not output. The engineer writes down every mistake Claude Code makes — those become rules in CLAUDE.md. By Friday, the file is two to three times longer than it was on Monday, and Claude Code is producing reliable output on the chosen track.

Stage 2 — Team rollout

The rest of the team installs. The Stage 1 engineer leads a one-hour walkthrough of the CLAUDE.md, the chosen MCP servers, and the team's review patterns. Every PR generated by Claude Code gets tagged in the description ([claude-code]) for the next month so the team can audit the actual quality.

Critical: the team's review standards do not change. A Claude Code PR is reviewed exactly like a human PR. The reviewer's job is to catch problems — the source of the code is not relevant.

Stage 3 — Specialist harnesses

Once the team is comfortable, we introduce specialist harnesses — narrow Claude Code configurations scoped to specific work patterns. A harness for the widget code looks different from a harness for the billing integration. Each harness has its own CLAUDE.md, its own MCP server set, and its own review rubric. This is where the productivity numbers start to look like the Forrester study.

The companion piece on what Claude Code can actually do for your business covers the specific use cases the harness pattern unlocks — for engineering teams and beyond.

Reviewing-model verification

For our own production work, every non-trivial Claude Code change passes through a second model — typically Opus 4.7 — for review before reaching a human. The reviewing model reads the proposed change with fresh context and grades it against a rubric: correctness, security, performance, style. The human reviews the reviewer's summary first, then the diff. Most regressions are caught here.

This pattern is how Cellbot — our flagship Claude-powered product for UK repair businesses — ships at the quality and pace it does. The full case study is at /case-studies/cellbot. The technical methodology paper goes up on Promptology later this month.

Frequently asked questions

Is Claude Code free to install?

The Claude Code CLI itself is free to install via npm. Using it requires an active Anthropic subscription — Pro at £17/month is the minimum. Free Anthropic accounts cannot run Claude Code.

Does Claude Code work on Windows natively?

Yes, since Claude Code v2 (March 2026). WSL is no longer required. Install via PowerShell 7+ with npm install -g @anthropic-ai/claude-code. Existing WSL installs continue to work; choose based on where the rest of your toolchain lives.

What Node version do I need for Claude Code?

Claude Code requires Node.js 18 or higher. We recommend Node 22 LTS, which is the current long-term support release as of 2026. Avoid using your operating system's package-manager Node — install via nvm (macOS/Linux) or fnm (Windows) so you can switch versions per project.

Can I install Claude Code without an Anthropic subscription?

You can install the CLI, but you cannot use it. The first time you run claude, the authentication flow checks your Anthropic plan. Free accounts are rejected. Sign up for Pro at claude.com/upgrade before installing.

How big a project can Claude Code handle?

Claude Code reads files on demand — it does not load your entire codebase into memory. Projects from a few files to several million lines work. The practical constraint is whether your CLAUDE.md is good enough to point Claude Code at the right files for any given task. For monorepos and very large projects, scoped sub-CLAUDE.md files in subdirectories solve most context-routing problems.

Does Claude Code work with monorepos?

Yes. Place a top-level CLAUDE.md describing the overall structure, then sub-CLAUDE.md files in each package or workspace describing the local conventions. Claude Code merges these from the deepest match upwards when reasoning about a task in a particular subdirectory.

Can I install multiple Claude Code instances on one machine?

Yes — Claude Code is a per-shell process. You can run multiple sessions in parallel against different projects. Each session reads its own CLAUDE.md and connects to its own MCP servers. The only shared resource is the OAuth credential, so all sessions run as the same user.

How do I update Claude Code?

`bash npm update -g @anthropic-ai/claude-code `

Anthropic ship updates roughly monthly. Major versions (v1 → v2) sometimes change CLI defaults — read the release notes before upgrading in production.

What happens if Anthropic deprecates a model I use?

Anthropic give six months notice before deprecating a model. Claude Code automatically falls back to the next-best available model in the same family if the configured model is unavailable. You can pin a specific model in your CLAUDE.md or via the --model flag if you need version stability for compliance reasons.

Is Claude Code safe for regulated UK businesses (finance, legal, healthcare)?

For SMEs in regulated sectors, the Pro plan with a properly configured CLAUDE.md and audit-friendly review process is usually enough. For larger organisations or strict compliance environments — FCA-regulated firms, NHS trusts, law firms handling sensitive client data — the Enterprise plan adds SSO, SCIM provisioning, audit logging, configurable data retention, and contractual zero-retention guarantees. We cover this in the companion piece on Claude Code for business.

Related reading

• ↑ What is Claude Code? A UK Business Guide — the foundational pillar this install guide sits under
• ↔ Claude Code Pricing 2026 — Real Cost for UK Businesses — what each plan actually costs once you map it onto a real engineering month
• ↔ Claude Code MCP Servers — 7 Worth Installing — the next file you should set up after CLAUDE.md
• ↔ Claude Code Skills — Write, Share, and Govern at Scale — the patterns for building reusable skills your whole team can use safely
• ↔ How to Deploy Hermes Agent — UK Business Complete Guide — the Hermes-side equivalent of this guide, for self-hosted ops automation that complements Claude Code

What should you do next?

The install takes five minutes. The first month of CLAUDE.md iteration is what actually changes how your team ships software.

If you want a pair of expert eyes on your rollout — a working session with a UK Claude Code practitioner who has shipped this in production, your CLAUDE.md reviewed against the patterns that actually scale, and a thirty-day plan tailored to your stack — book a free Claude Code working session.

Book a free Claude Code working session →

Forty-five minutes, free, no commitment. We cover your install, your CLAUDE.md, the right MCP servers for your stack, and the rollout sequence that gets you to the productivity numbers Anthropic talk about. You leave with a plan you can execute that afternoon.