How to Use Claude Code¶

VSCode¶

https://code.visualstudio.com/download - select appropirate one Windows/Mac/Linux. Enusre 'Add to path' checkbox selected but other than that accept defaults.

Install Git¶

https://git-scm.com/install/windows - accept all defaults.

Install Node¶

Node.js needs installing as Claude Code is built in TypeScript/JavaScript:

https://nodejs.org/en/download

Python does not need to be installed unless one is using Python projects. This may be important for Data Scientists/Analysts.

Install UV¶

Optional but the Data Intelligence Researcher Agent uses this but can be done with traditonal 'pip'.

The modern UV installer will install it for you in your project if you have issues.

Install UV here: https://docs.astral.sh/uv/getting-started/installation/

Set Up Agent¶

git clone https://github.com/Python-Test-Engineer/data-intelligence-researcher in a folder of your choice then open up that folder in VSCode.

What Is Claude Code?¶

Claude Code is a tool you run in your computer's terminal (the black window with text). You type instructions in plain English, and it reads your files, writes code, browses the web, and carries out tasks — all on your computer, in your project folder.

Think of it as a very capable assistant who lives in your terminal and never gets tired.

You do not need to be a programmer to use it. If you can describe what you want in a sentence, Claude Code can usually do it.

Before You Start¶

You need three things:

Node.js installed on your computer (nodejs.org — download the LTS, Long Term Support version)
An Anthropic API key (sign up at console.anthropic.com)
A terminal — on Windows, use PowerShell or Windows Terminal; on Mac, use Terminal

Installation¶

Open your terminal and run:

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

Starting Claude Code¶

Navigate to your project folder in the terminal:

cd /path/to/your/project

Then start Claude Code:

claude

You will be prompted for your Subscription or API KEY but ensure you have already installed Node.

You will see a prompt. Type your instruction and press Enter.

> Summarise what this project does based on the files in this folder

That is it. Claude Code reads your files and responds.

Your First Five Tasks¶

Here are five things to try on your first day. Each one takes less than a minute.

1. Summarise your project

What files are in this folder and what does this project do?

2. Read and explain a file

Read README.md and explain it in plain English

3. Create a new file

Create a file called notes.md with a section for ideas and a section for tasks

4. Edit an existing file

Add a line to notes.md under Tasks: "Set up Claude Code"

5. Search for something

Find all files in this project that mention the word "invoice"

How to Give Good Instructions¶

Claude Code works best when your instructions are clear and specific. Think of it like giving directions to someone who is very capable but does not know your situation unless you tell them.

Too vague:

Fix the website

Clear and specific:

Read docs/01-website/intro.md and fix any spelling mistakes. Do not change the meaning or structure.

Even better (tells it where to save):

Read docs/01-website/intro.md. Fix any spelling mistakes and save the corrected version to the same file.

Tips for better instructions¶

Say what file to read — Claude Code can find files, but pointing it directly is faster
Say what format you want the output in — table, bullet list, paragraph, markdown
Say where to save the result — "save to /output/summary.md"
Give it a length target — "200 words", "one page", "under 10 bullet points"
Tell it the audience — "explain this for a non-technical business owner"

Modes of Working¶

Interactive mode (default)¶

You type an instruction, Claude Code responds, you type another. Good for exploratory work, editing, and quick tasks.

> Read my CV at /docs/craig-cv.pdf and list my top 5 skills

One-shot from the command line¶

Run a task without entering interactive mode. Good for scripts and scheduled tasks.

claude "Read /analytics/this-week.csv and write a summary to /analytics/summary.md"

Understanding What Claude Code Can See¶

By default, Claude Code can see files in the folder you started it from (your working directory). It can read subfolders too.

It cannot see files outside that folder unless you install the filesystem MCP, (Model Context Protocol), server (covered below) or give it an absolute file path.

It cannot browse the internet by default — you need a web search MCP for that.

It cannot send emails, post to social media, or interact with external services unless you give it the right tools.

The CLAUDE.md File — Your Standing Instructions¶

CLAUDE.md is a special file you create in your project root. Claude Code reads it automatically at the start of every session. Use it to give Claude Code permanent instructions about how to behave in your project.

Create it:

> Create a CLAUDE.md file in this folder with instructions for working on this project

Or write it yourself:

# My Project — Claude Code Instructions

## Rules
- Always write in British English
- Keep file names lowercase with hyphens (not underscores)
- Never delete files — move them to /archive/ instead
- Ask before making changes to mkdocs.yml

## Key files
- /docs/index.md — main homepage
- /docs/clients.md — client list (do not share publicly)
- /templates/ — document templates to use for all new docs

## Writing style
- Tone: direct and practical
- No corporate jargon
- Audience: solo business operators

Every time you start Claude Code in this project, it will follow these rules without you having to repeat them.

MCP Servers — Extending Claude Code's Powers¶

MCP (Model Context Protocol) servers add new capabilities to Claude Code. They are like plugins.

Installing an MCP server¶

claude mcp add <server-name>

The most useful MCP servers for beginners¶

MCP Server	What It Adds	Install Command
`filesystem`	Read/write files anywhere on your computer	`claude mcp add filesystem`
`fetch`	Retrieve content from any web URL	`claude mcp add fetch`
`brave-search`	Search the web in real time	Requires a Brave API key
`puppeteer`	Control a web browser, take screenshots	`claude mcp add puppeteer`
`sqlite`	Query and update a local database	`claude mcp add sqlite`

Example: adding web search¶

Once you have a Brave Search API key:

claude mcp add brave-search

Then in Claude Code:

> Search the web for the latest news about AI productivity tools and summarise the top 3 results

Practical Workflows for Beginners¶

Workflow 1: Turn rough notes into a polished document¶

Write your rough notes in a text file (bullet points, fragments, anything)
Save it as /drafts/rough-notes.txt
Run:

Read /drafts/rough-notes.txt.
Turn these rough notes into a polished 500-word article.
Tone: practical and direct. Audience: small business owners.
Save to /drafts/polished-article.md

Workflow 2: Summarise a long document¶

Read /documents/long-report.pdf.
Produce a one-page executive summary with:
- The 5 key findings
- 3 recommended actions
- Any risks or concerns mentioned
Save to /documents/long-report-summary.md

Workflow 3: Organise a messy folder¶

List all files in /downloads/.
Group them by type (PDFs, images, spreadsheets, documents).
Suggest a folder structure I could use to organise them.

Workflow 4: Write a professional email¶

Draft a professional email to a client.
Context: they asked for a project update, the project is on track,
we deliver the first version next Friday.
Tone: confident and reassuring. Length: 100 words.

Common Mistakes¶

Mistake: Instructions that are too open-ended

"Make the website better"

Claude Code will try, but it does not know what "better" means to you. Be specific about what you want changed.

Mistake: Forgetting to say where to save output

If you ask Claude Code to write something and do not say where to save it, it will print it to the screen. Add "save to /folder/filename.md" to every instruction that produces a document.

Mistake: Not reviewing before using

Claude Code is very capable but not perfect. Always read its output before sending an email, publishing a page, or sharing a document. It can make factual errors, especially about things that have happened recently.

Mistake: Trying to do everything in one instruction

Break complex tasks into steps. One clear instruction at a time produces better results than one long instruction that tries to do ten things.

Keyboard Shortcuts¶

Shortcut	What It Does
`Ctrl + C`	Stop the current task
`↑` arrow key	Recall previous instruction
`/clear`	Clear the conversation history (fresh start)
`/help`	Show available commands
`/model`	Switch between Claude models

Getting Help¶

If Claude Code does something unexpected:

Ask it to explain: What did you just do and why?
Ask it to undo: Revert the last change to that file
Start fresh: Type /clear to reset the conversation
Check the file: Use Read or open the file directly to see what changed

If you are stuck, describe what you were trying to do and what happened:

I asked you to edit /docs/services.md but the file looks wrong now.
Can you show me what the file contains and explain what changed?

Next Steps¶

Once you are comfortable with the basics:

Set up your CLAUDE.md with rules specific to your business
Install the fetch MCP to pull in web content
Try a multi-step workflow (research → write → save)
Explore the other sections of this manual for business-specific workflows

The more you use Claude Code, the better you will get at writing instructions that get exactly what you want on the first try.