How Claude Code Sees Files | Claude Code Guide

When you start Claude Code from a folder, that folder becomes its world. Everything inside it is visible. Everything outside it doesn't exist. This is the single most important thing to understand about how Claude Code works. Get this right, and everything else makes sense. Get it wrong, and you'll spend half your time wondering why Claude can't find things. ## The Room Analogy Think of your computer as a house with many rooms. Each room is a folder. Documents, Downloads, Desktop, Projects. Each room has stuff in it. When you open Claude Code, you're choosing which room to let it into. Whatever folder your terminal is in when you type `claude`, that's the room Claude sees. Let's use the practice folder we created earlier. Open your terminal and navigate to it: **Windows (PowerShell):** ```powershell cd "$HOME\Documents\practice" ``` **Mac/Linux:** ```bash cd ~/Documents/practice ``` Then start Claude Code: ```bash claude ``` Now Claude sees everything in `practice`. Every file, every subfolder, every piece of text inside those files. But it can't see your Desktop. It can't see your Downloads. It's in the project room, and that's where it stays by default. (Technically, Claude *can* access files outside the project folder if you explicitly point it there and grant permission. But for most workflows, you want Claude focused on one project at a time. We'll stick to that approach in this guide.) This is a good thing. You want Claude focused on one project at a time. You wouldn't hand someone every document you own and say "help me with my website." You'd hand them the website folder. **Pro tip:** If you have a file you want Claude to work with - say, a downloaded image or a document from your email - copy it into your project folder first. Don't try to reference files scattered across Downloads or Desktop. Keep everything Claude needs in one place. ## Five Things Claude Can Do With Your Files Once Claude is in your project folder, it can: Read files "What's in index.html?" or "Show me the contents of notes.txt." Claude opens the file and reads it. All of it, including files inside subfolders. Create files "Create a file called about.html with an about page." Claude writes the file from scratch and puts it in your project folder. Edit files "In style.css, change the background color to dark blue." Claude finds the file, finds the specific part, and changes it. Organize and rename "Rename all the .jpeg files to use .jpg instead" or "Move all images into a folder called assets." Claude rearranges things. Search across files "Find every file that mentions the word 'invoice'" or "Which CSS file controls the font size?" Claude searches through everything and tells you what it found. All of this happens because you asked in English. No commands to memorize. ## The @ Shortcut Sometimes you want to point Claude at a specific file. Instead of saying "look at the file called styles/main.css," you can use the @ symbol: ```text Look at @styles/main.css and tell me what colors are defined ``` The @ shortcut is just a faster way to reference files. Claude will find them either way, but @ makes it explicit. Especially useful when you have many files with similar names. ## Beyond Single Files: Working With Groups It gets more interesting. The @ symbol isn't just for pointing at one file. You can reference entire folders: ```text Look at everything in @notes/ and tell me what topics come up most often ``` That trailing slash matters. It tells Claude "this is a folder, read everything inside." Claude will go through every file in that folder and look for patterns, contradictions, themes, or whatever you ask about. This opens up a few useful workflows: **Pulling insights from scattered files.** You have a folder of meeting notes, customer feedback, or research documents. Instead of reading them one by one, ask Claude to find the common threads. "What are people complaining about most?" "What requests keep coming up?" "Where do these documents contradict each other?" **Applying a consistent format.** You have raw content and a template you like. Point Claude at both: "Rewrite @raw-notes.md using the structure from @my-template.md." The template becomes a reusable pattern. **Comparing things.** "Read @proposal-v1.md and @proposal-v2.md and tell me what changed." Or point at a folder of competitor websites you've saved and ask for a comparison. The key insight: Claude can hold multiple files in mind at once. That means cross-referencing, comparing, and synthesizing are all things you can ask for in plain English. ## Showing Claude Images Something most people don't realize: Claude Code can see images too. If you have a screenshot you want Claude to look at, press **Ctrl + V** (or **Cmd + V** on Mac) to paste it directly into the conversation. Claude will describe what it sees and can work with it. This is wildly useful for design work. Take a screenshot of a website you like, paste it in, and say "build something that looks like this." Claude sees the layout, the colors, the structure. It's not perfect, but it's a remarkable starting point. You can also reference image files in your project: ```text Look at @mockup.png and build an HTML page that matches this design ``` Beyond design, images work for: - Screenshots of error messages (faster than typing them out) - Charts and graphs you want analyzed - Handwritten notes you want transcribed - UI mockups you want turned into code - Photos of whiteboards from brainstorming sessions Visual input often communicates what words struggle to describe. ## Finding Current Information Claude Code can also search the web when you need current information: ```text What's the recommended way to handle form validation in 2024? ``` This is useful when you're building something and want to follow current best practices, or when you need to reference something that might have changed since Claude's training data. You can combine this with your files: "Search for modern approaches to responsive navigation, then update @nav.css to follow those patterns." ## Where Claude Puts Things When you ask Claude to create a file, it puts it in the folder where you started Claude Code. If you want files somewhere specific, say so: - "Create `pages/about.html`" puts it in a `pages` subfolder - "Create `about.html`" puts it in the project root - "Create a `css` folder and put `styles.css` inside it" creates the folder and file Be specific about where things should go, especially as your project grows. "Create a file" is fine for practice. "Create `pages/contact.html`" is better for real work. ## What Claude Remembers Within a single conversation, Claude remembers everything it's seen and done. This is what people call "context" - that buzzword you've probably heard about AI models. More context means the AI can keep track of more information at once. It read your HTML file? It remembers the contents. It created a CSS file? It knows what's in it. That's all part of the context. But the marketing hype doesn't mention something important: bigger context isn't always better. A massive context window filled with irrelevant information can actually make Claude less focused. So in later chapters, we'll cover memory files and rules, ways to give Claude persistent knowledge without cluttering every conversation. When you exit and start a new conversation, Claude starts fresh. It can see your files again (they're still there), but it doesn't remember the conversation you had. It will re-read files as needed. This is why the folder you start from matters so much. Your files are the bridge between sessions. Claude's memory is the conversation. Your files are the permanent record. ## Real-World Example My website has over 200 files spread across 20-something folders. CSS files, PHP files, JavaScript, images, documentation. When I need something, I'm specific: "Look at `public/css/components/nav.css` and fix the mobile menu animation." Claude goes straight to that file, reads it, understands it, and makes the change. But I can also go broad: "Search all CSS files for anything using the color green." Claude scans every CSS file and gives me a list. That kind of search across 200 files would take me ages manually. Claude does it in seconds. Claude sees everything in the project. You decide whether to point it at one file or let it search through all of them. > Next up: how to give Claude Code better instructions. The difference between vague and specific changes everything.

How Claude Code Sees Your Files

The Room Analogy

Sign In to Unlock

Check Your Email