Claude Code setup: a practical guide for UK small teams

Claude Code is Anthropic’s coding agent. It reads the files in a folder, edits them and runs commands for you, from your terminal, your code editor or the Claude desktop app. If your team already uses Claude for writing and analysis, it is the next step for anyone whose work lives in files, from a full-time developer to the marketer who looks after the website and the ad exports.

Below is how a small team should set it up, including the defaults Anthropic changed in September and October 2026. We checked every version, default, command and price against Anthropic’s documentation, GitHub release notes and pricing page on 16 October 2026. We also installed Claude Code 2.1.301 in Ubuntu under WSL on a Windows PC and had it build a small marketing tool; the screens below come from that run. Prices are US list prices and exclude tax.

The answer in brief

Sign up for a paid Claude plan, then install Claude Code with Anthropic’s native installer, or open the Code tab in the Claude desktop app if you would rather avoid the terminal. Before anyone works on real files, commit a short .claude/settings.json with deny rules for secrets and a CLAUDE.md of house rules to each project, and decide whether to keep auto mode, which is now the default in terminal and VS Code sessions. Use a Team plan for shared client work, and have marketers keep research and documents in the main Claude app and bring file, data and website jobs to Claude Code.

What is Claude Code, and which plan do you need?

Claude Code is an agent: you describe a task in plain English and it works through the files in a folder, proposing and making changes and running tests or other commands as it goes. It needs a paid account. Anthropic’s setup guide says it requires a Pro, Max, Team, Enterprise or Console account, and that the free plan does not include it. The UK is on Anthropic’s list of supported countries for both the Claude apps and the API.

Source: Claude pricing and Claude Code setup, checked 16 October 2026. These are Anthropic’s US dollar list prices, excluding tax. The pricing page can show prices in a local currency depending on where you view it from, so check the price and any VAT at checkout.

For a business sharing client work, we would choose Team over a handful of personal Pro accounts, even at a slightly higher price per head. Under commercial terms Anthropic does not train on code sent through Claude Code, and Team gives you an admin console for spend limits and managed settings. A sole developer working on their own code can start on Pro.

How do you install Claude Code?

There are two routes. The desktop app suits anyone who doesn’t use a terminal; the native installer suits developers. Both run the same Claude Code and read the same settings files.

In the desktop app, without a terminal

Anthropic’s desktop quickstart says the app runs Claude Code “without using a terminal”, and you don’t need to install Node.js or the command-line tool.

1. Download and sign in. Install the Claude desktop app for macOS or Windows (Linux is in beta) and sign in with your paid Claude account.
2. Open the Code tab. If it asks you to upgrade, your account isn’t on a paid plan.
3. Choose a folder. Select Local, then Select folder, and pick the project. Anthropic’s tip is to “start with a small project you know well”; we would work on a copy of the live website.
4. Choose a permission mode. Use the selector next to the send button. Manual shows each change as a diff, the line-by-line before and after view, and waits for you to accept it.
5. Try a safe first task. Ask: “List the files in this folder and explain what each one does. Don’t change anything.” Then try one small, reversible edit.

The app remembers the mode you pick for each folder, apart from Plan, which lasts only for the current session, and that choice overrides your settings for that folder. To make new local sessions in other folders start in Manual too, add "permissions": {"defaultMode": "default"} to the user settings file covered below; default is the settings name for Manual.

In the terminal

Claude Code runs on macOS 13 or later, Windows 10 version 1809 or later, Windows Server 2019 or later, Ubuntu 20.04 or later, Debian 10 or later and Alpine Linux 3.19 or later, with at least 4 GB of RAM (setup guide). Anthropic recommends its native installer:

• macOS, Linux and WSL: curl -fsSL https://claude.ai/install.sh | bash
• Windows PowerShell: irm https://claude.ai/install.ps1 | iex

Open a new terminal and run claude --version; a version number means it worked. Then run claude in a project folder and sign in through the browser. claude doctor prints a read-only health check of the install and your settings files. On our Ubuntu install under WSL, it reported version 2.1.301 as a native install on the latest update channel, with no installation issues. If your terminal says claude isn’t found, Anthropic’s installation troubleshooting page explains how to fix your PATH.

VS Code users can instead install the Claude Code extension: open the Extensions view, search for “Claude Code” and click Install (VS Code guide). The extension works with any paid Claude plan or a Console account.

On Windows, install Git for Windows first so Claude Code can use Bash; without it, Claude Code runs commands through PowerShell. If you want sandboxing, which isolates the commands Claude runs from the rest of the machine, use WSL 2: the setup guide says native Windows does not support it.

Which update channel should a team use?

Pin everyone to the stable channel. Native installs update themselves in the background, and the default latest channel takes each release as it ships. The stable channel is “typically about one week old” and skips releases with major regressions. Choose it in /config under Auto-update channel, or add "autoUpdatesChannel": "stable" to your user settings file. With releases most days, a week’s delay costs little and keeps the team off releases that turn out to have major regressions. The exception is new models: Anthropic says a newly launched model can need a newer version than the stable channel serves, so switch to latest if you need one on launch day. Homebrew users choose through the cask name: claude-code tracks stable and claude-code@latest tracks latest.

Should your marketing team use Claude Code or Claude?

Use Claude for most marketing work, and Claude Code for jobs that live in files, data or a website. Anthropic draws the line itself. Its guide to the desktop app says “Code is for building software”, while Cowork, Claude’s background agent for delegated work, is for “complex, multi-step work you delegate”, such as research across many sources and finished documents. Anthropic is now merging Cowork into the main Claude app, starting with Pro and Max. When it launched workflows for small businesses on 15 September 2026, including drafting next week’s social posts and logging new leads to your CRM, it built them for Cowork.

Claude Code is the better tool when a marketing job means working through files. Anthropic’s prompt library for Claude Code tags several prompts for marketers, including one that reads an ad performance export, finds the weakest headlines and writes 20 new ones under 90 characters, and another that finds every version of a phrase across a project and replaces it. In January 2026, Anthropic described how a growth marketer who had never written code built a Figma plugin for ad variants and a workflow that turns campaign data into Google Ads responsive search ad copy and exports it as a CSV. By Anthropic’s own count, each ad went from 30 minutes to 30 seconds.

We tried a job of that kind ourselves. Given an export of eight made-up ads for a furniture shop, Claude Code built a one-page headline checker from one plain-English request: open the CSV, rank the headlines by click-through rate and flag any longer than 30 characters, with no external libraries. It wrote the page, checked its sums against the file with a short script and finished in 49 seconds. It also said plainly that it had not opened the page in a browser. When we did, the page worked, and it flagged six of the eight headlines as too long, including the two best performers.

Source: AIWIZ view based on Anthropic’s desktop app guide, its Claude for Small Business announcement and the Claude Code prompt library, checked 16 October 2026.

One addition from us: use Manual mode for any session that can reach a live ad account. Auto mode’s default block list targets software and data risks such as force pushes and production deploys, and names no ad budget or campaign actions. Whatever the tool, check AI-written copy before it goes live; our post on Google’s AI content guidance sets out what to check.

Which settings files should a small team commit?

Claude Code reads settings from four places, and the one you choose decides who a setting applies to (settings guide). Your user file, ~/.claude/settings.json, covers you in every project. A project’s .claude/settings.json covers everyone who works in that folder once you commit it, which means saving it to the team’s shared Git repository. .claude/settings.local.json holds your personal overrides for one project and stays out of Git when Claude Code creates it. Managed settings come from your organisation and override all of them. The terminal, the VS Code and JetBrains extensions and the desktop app all read the same files. If editing JSON is new to you, you can ask Claude Code to create or change these files and review its edit in the diff.

The shared project file is where a small team gets most value. A starting point, adapted from Anthropic’s own example:

{
  "$schema": "https://json.schemastore.org/claude-code-settings.json",
  "permissions": {
    "allow": [
      "Bash(npm run lint)",
      "Bash(npm run test *)"
    ],
    "deny": [
      "Read(./.env)",
      "Read(./.env.*)"
    ]
  }
}

The allow rules let Claude run your lint and test commands without asking. The deny rules block Claude’s file tools, and shell commands such as cat that name the file, from reading environment files that usually hold API keys and passwords. Swap in your own build commands, and add a deny rule for any file that holds ad platform, CRM or analytics credentials. In our test project, asked to show the contents of .env, Claude Code refused, said the permission settings blocked it and suggested how to view the file yourself.

Know the limits. Anthropic’s permissions guide says deny rules don’t cover a command that reads files without naming them, such as a search across the whole folder, or a script that opens files itself; for that it points to the sandbox, which native Windows doesn’t support. The safest place for a live credential is outside the project folder. Within those limits, deny rules win: Claude Code checks deny rules, then ask rules, then allow rules, so a narrow allow rule cannot carve an exception out of a broad deny rule. Outside Team and Enterprise plans, on a machine without managed settings, an installed mod can override them, as the mods section explains.

Committed allow rules also wait for trust. The first time anyone opens Claude Code in a folder, it asks whether they trust it, and the highlighted answer is “No, exit”. Until a teammate says yes, the project’s allow rules don’t apply for them, while deny and ask rules apply at once. Run /status in a session to see which settings files loaded.

What should go in CLAUDE.md?

CLAUDE.md is a plain Markdown file of standing instructions that Claude reads at the start of every session (memory guide). Run /init in a project and Claude drafts one from your files, with build and test commands and the conventions it finds. Then trim it.

Anthropic’s guidance is specific: aim for under 200 lines, and write instructions concrete enough to check, such as “Run npm test before committing” instead of “Test your changes”. Add to it when Claude makes the same mistake twice or a reviewer catches something Claude should have known. Personal notes, such as your own test URLs, belong in CLAUDE.local.md, which you add to .gitignore. For a marketing project, the same file holds your brand’s spelling rules, banned claims and character limits.

Two autumn changes help. Since version 2.1.277, Claude Code reads an existing AGENTS.md when a project has no CLAUDE.md or CLAUDE.local.md, so you needn’t copy instructions you already keep for another coding agent. Adding a CLAUDE.local.md switches that off; to keep both, set Project instructions in /config to read CLAUDE.md and AGENTS.md together. Since 2.1.283, /doctor prompt-audit reports instructions written for older models, references to files or commands that don’t exist and files that contradict each other, and changes nothing until you ask.

Claude treats CLAUDE.md as context and can still ignore a vague or conflicting line. Anything that must never happen belongs in a deny rule, a hook or the sandbox.

What does auto mode mean for your team?

Auto mode lets Claude work without stopping for routine approvals. Anthropic’s permissions guide describes it as running “without routine prompts”, with a background classifier checking that shell commands and network requests match what you asked for. In Manual mode, by contrast, Claude asks before most edits, commands and web requests.

The classifier is stricter than a “yes to everything” switch. Anthropic’s permission modes guide lists what it blocks by default, including downloading and running code with curl | bash, sending sensitive data to outside endpoints, production deploys and migrations, force pushes, git reset --hard, changes to DNS records or secret managers, and merging a pull request that no person has approved.

What changed is who gets it without asking. Release 2.1.284 on 28 September 2026 made interactive terminal and VS Code sessions start in auto mode “on every plan and provider” whenever no permission mode is configured (release notes); Anthropic’s permission modes guide dates the same change to 2.1.283. Before 2.1.283 on 25 September, only Pro, Max and Team sessions started there by default; that release added third-party providers and sessions with telemetry off. Anyone who opens Claude Code in a terminal or VS Code today, with no permission mode set and a supported model, starts in auto mode unless their organisation has turned it off.

The software says so too. When we first started version 2.1.301, its welcome screen read “Auto mode is now Claude Code’s default permission mode.” During our build it wrote a 252-line file without asking, then ran a check command that the screen marked “Allowed by auto mode classifier”.

For most coding work we would keep that default. Three settings deserve a decision first:

• Where the mode is set. Claude Code ignores bypassPermissions in a project’s .claude/settings.json and starts in Manual mode instead, so a cloned repository cannot turn your prompts off. It ignores auto there too, and then falls back to the built-in default, now auto mode, rather than the default in your own settings file. Otherwise a defaultMode in project or managed settings outranks your user file, ~/.claude/settings.json; use --permission-mode to choose for one session.
• Who can switch it off. On Team and Enterprise plans an administrator can remove auto mode, or bypass mode, for everyone by setting permissions.disableAutoMode or permissions.disableBypassPermissionsMode to "disable" in managed settings.
• Which work needs Manual mode. For sessions that hold production credentials, touch client infrastructure or reach a live ad account, start in Manual mode, and change mode (Shift+Tab in the terminal, the mode selector in the desktop app) only when you mean to.

In Manual mode each edit arrives as a diff with three choices: yes; yes, and accept edits for the rest of the session; or no. Choosing no leaves the file as it was.

Bypass permissions mode skips permission prompts altogether. Anthropic says to use it only in isolated environments such as containers or virtual machines, and we would not let it near a laptop with real credentials.

Which model should the team start on?

On Pro plans and Team Standard seats, Claude Code now starts on Opus. Release 2.1.280 changed the default “from Sonnet to Opus, matching Max, Team Premium, and Enterprise” on 22 September 2026, the day Claude Opus 5.5 became the default Opus model.

Opus suits complex work and costs more than routine jobs need. Anthropic’s cost guide says Sonnet “handles most coding tasks well and costs less than Opus”, and suggests keeping Opus for complex architecture and multi-step reasoning. Anthropic’s pricing page lists Opus 5.5 at $4 per million input tokens and $20 per million output, and Sonnet 5.5 at $2 and $10. Switch with /model. For quick subagent jobs, Anthropic suggests model: haiku, which since version 2.1.293 on 7 October 2026 means Claude Haiku 5.5 on the Anthropic API.

Effort is the other setting to know. Opus 5.5, Sonnet 5.5 and Haiku 5.5 all start at medium effort in Claude Code; our prompting guide covers the /effort command and when to change it.

What will Claude Code cost a small team?

On a Team plan, the seat price is the main cost. Each member’s Claude Code use draws on a per-seat allowance that resets on “a rolling five-hour window and a weekly window”, shared with Claude chat and Cowork, and its size depends on whether the seat is Standard or Premium (cost guide). When a member runs out, they wait for the reset, or an admin turns on usage credits and sets spend limits for the organisation, a group or a person.

If you pay per token instead, Anthropic’s figure across enterprise deployments is “around $13 per developer per active day”, or $150 to $250 per developer per month, with 90 per cent of users below $30 per active day. Use it as a planning figure, and run a small pilot to set your own baseline before rolling out.

Don’t count on the API credits that Max and Team plans now include each month: Anthropic’s credits page says they don’t cover Claude Code.

On per-token billing, Anthropic says unexpectedly high spend usually traces back to long sessions that were never cleared, or to Opus left as the default model. Claude Code sends the whole conversation with every request; prompt caching makes repeat history cheaper, but you still pay for it. Clear it with /clear when you switch to unrelated work, and check /usage to see what is drawing on your allowance.

What happens to your code and data?

It depends on the plan, which is why we prefer Team for client work (data usage). On Team, Enterprise and the API, Anthropic does not train its models on code or prompts sent to Claude Code under commercial terms, unless your organisation chooses to share data for model improvement, and it keeps data for 30 days as standard. Pro and Max use consumer terms: Anthropic trains on your data when the model-improvement setting is on, and keeps it for five years if you allow that or 30 days if you don’t. Anyone on a personal Pro account should check that setting before opening a client’s files.

Claude Code also keeps session transcripts on each machine, in plain text under ~/.claude/projects/, for 30 days by default. Transcripts of sessions started or most recently continued in the desktop app or Cowork are exempt from that limit by default, so treat laptops accordingly. If you need processing to stay in Europe, our Claude Haiku 5.5 post sets out the cloud routes and their trade-offs.

Should you install mods?

Only from sources you would trust with your machine. Anthropic introduced mods on 1 October 2026: short TypeScript add-ons, packaged in plugins, that alter Claude Code’s behaviour (Anthropic’s mods announcement). They can rewrite prompts, block or rewrite tool calls, approve or deny permission requests and add interface elements.

Anthropic says mods run with the same access to your machine as Claude Code and are not sandboxed. On Team and Enterprise plans, or on machines with managed settings, deny rules hold over a mod by default, and a built-in mod called sec-default loads first to block risky moves. Anywhere else, Anthropic’s permissions guide says a mod “can approve a call that a deny rule refuses”, which is one more reason to run a team on a Team plan. Whatever your plan, treat a mod like any other code you install, and read it first. The same caution applies to the marketing plugins and skill packs now shared online.

What changed in Claude Code this autumn?

Anthropic releases new 2.1.x versions most days, and most are fixes. These changed what a new team sees on day one.

Source: Claude Code releases on GitHub and, for mods, Anthropic’s mods announcement, checked 16 October 2026. The newest release that day was 2.1.301, published on 15 October 2026.

What should a small team do this week?

One person can do most of this in a week, starting with the plan and ending with a look at the first fortnight’s usage.

1. Choose the plan. Move anyone using Claude Code on client work to Team seats, and decide who needs Premium.
2. Decide which jobs go where. Keep research, documents and social posts in the main Claude app, and send file, data and website work to Claude Code.
3. Install and check. Use the desktop app or the native installer. On command-line installs, set autoUpdatesChannel to stable and run claude doctor.
4. Commit a shared settings file. Add deny rules for secrets and allow rules for your test and lint commands to each project’s .claude/settings.json, and keep live credentials outside project folders.
5. Write a short CLAUDE.md. Start with /init, cut it below 200 lines, and run /doctor prompt-audit once a month.
6. Decide on auto mode. Keep it for normal development, use Manual mode for anything with production or ad account access, and have an admin set the policy in managed settings.
7. Watch the first fortnight’s usage. Check /usage and the admin spend report, and move routine work from Opus to Sonnet if allowances run short.

If you want help rolling Claude Code out across a team or wiring it into your marketing workflows, see our AI adoption services, our web development team or talk to AIWIZ.