Uploading Documents and PDFs

One of the coolest features of the Claude web interface is the ability to upload files directly. Claude doesn't just read the text; it can parse PDFs, CSVs, markdown files, and standard code files.

The Attachment Button

In the chat interface, there's a small paperclip icon. Clicking this lets you upload up to 5 files at a time (on the standard UI).

Why is this useful?

Imagine you just bought a new smart home device, and the API documentation is a 50-page PDF. Instead of manually searching through it to figure out how to turn a lightbulb on via a curl request, you just upload the PDF.

Prompt:

"I uploaded the API documentation for my smart home hub. I need to write a Node.js script that authenticates with the hub and turns the living room light (Device ID: 123) to blue. Give me the code."

Claude will read the entire 50-page PDF, find the authentication flow, find the specific endpoint for color changing, and write the script for you.

Analyzing Data (CSVs)

Claude isn't a replacement for a data scientist running Python scripts, but for quick and dirty data analysis, it's incredible.

If you upload a CSV of last month's sales data, you can ask plain-language questions:

"Look at the attached CSV. Which product category had the highest refund rate, and is there any correlation with a specific day of the week?"

Claude will parse the raw data and give you an answer. Note: It's doing this via natural language reasoning, not by executing SQL queries under the hood (unless you are using advanced tools, which we'll cover later). Always double-check its math if the dataset is massive, but for quick insights, it's a huge time saver.

Context Limitations

Even though Claude has a massive context window, uploading files does consume that context. If you upload three 100-page PDFs, every single time you ask a follow-up question in that chat, Claude has to re-process all 300 pages.

This can lead to slower response times. If you only care about Chapter 4 of a PDF, it's often better to just extract Chapter 4 and upload that, rather than making the AI chew on the whole book every time you speak.