Upload files to Claude two ways: click the plus icon next to the chat input and choose Add files or photos, or drag files directly onto the chat window. Claude accepts PDFs (multimodal under 100 pages, text-only over 1000), CSVs, TXT, DOCX, XLSX, and common image formats up to 500MB per file and 20 files per chat at 8000x8000 pixels max. Project files cap at 30MB each but the file count is unlimited as long as the total fits Claude's context window. Once attached, prompt Claude in plain English (summarize the PDF, calculate a ratio across the CSV, describe the image) and it reads or analyzes the file inline. Free on every Claude plan. Verified June 2026.

How to Upload Files to Claude: PDFs, CSVs, and Images (2026)

Claude's file upload is one of its most underused features in 2026: it handles three file modes in one chat (multimodal PDFs, structured data like CSVs and XLSX, and images), it reads visual elements inside PDFs under 100 pages instead of falling back to text extraction, and it works on every plan including Free. The bottleneck for most users is knowing the limits (500MB per file, 20 files per chat, 8000x8000 image max) and writing the right prompt afterward. This guide walks the full upload flow with screenshots from June 2026: opening the plus menu, dragging files onto the chat, managing attachments in the Files panel, and then writing the prompts that actually pull useful work out of an uploaded PDF or CSV. It also shows where the Claude module of AI Toolbox (formerly Claude Toolbox) picks up: full-text search across every Claude conversation so you can find the chat where you uploaded that quarterly report, message-level bookmarks for important responses, and per-conversation TXT or JSON export. AI Toolbox is part of the same unified Chrome install that powers our ChatGPT and Gemini modules.

What Files Can You Upload to Claude in 2026?

Claude accepts PDFs, CSV, TXT, DOCX, XLSX, PPTX, and common image formats (JPG, PNG, GIF, WEBP) up to 500MB per file and 20 files per chat. Image dimensions cap at 8000x8000 pixels; Anthropic recommends at least 1000x1000 pixels for best results because small or low-resolution images degrade analysis quality. PDFs under 100 pages run in multimodal mode (Claude analyzes both text and visual elements like charts and graphics); PDFs over 1000 pages drop to text-only extraction. Between 100 and 1000 pages, multimodal is best-effort. Page numbers in prompts should reference the PDF viewer's page count, not the page numbers printed on the document itself.

Project files have stricter per-file limits but looser per-project limits: 30MB per file, unlimited files per project, total content must fit within Claude's context window. Project file content is text extraction only except for multimodal PDFs, so a PNG inside a Project will not be read visually. For visual analysis, attach images directly to a chat, not the Project.

Step 1: Open the Plus Menu and Click Add Files or Photos

Click the plus icon (+) on the left of the chat input to open the upload menu, then click the top entry: Add files or photos (keyboard shortcut Cmd+U on Mac, Ctrl+U on Windows). The plus menu shows nine entries in 2026: Add files or photos (the upload action with paperclip icon and ⌘U shortcut), Take a screenshot (capture from your screen on desktop), Add to project, a separator, Skills (Anthropic's new repeatable-behavior surface that's absorbing Styles), Add connectors, Add plugins..., another separator, then Research, Web search (with a blue checkmark when enabled), and Use style. Click Add files or photos to open the native file picker. You can multi-select files of different types in one go (Cmd+click or Shift+click) and Claude attaches them all to the current prompt.

Take a screenshot is the Mac-only shortcut to capture the visible screen and attach it as an image without leaving the chat. Useful for screenshotting an error message, a chart in another app, or a tooltip that the screen-recorder permission lets Claude see directly. Both routes end up in the same place: attached files preview above the prompt input, ready to send.

Step 2: Or Drag Files Directly Onto the Chat Window

Drag any file from Finder, the macOS Desktop, or any other window onto the Claude chat window and a Drop files here to add to chat overlay appears with a dotted file icon in the center. Release the drag and Claude attaches the file the same way the plus menu does. Drag-and-drop accepts the same file types and limits as the menu route (PDFs, CSVs, DOCX, XLSX, images, text files, 500MB per file, 20 per chat). Multi-file drags work the same way: select multiple files in Finder, drag the group onto the chat, drop. Files attach in the order they were selected; reorder them later from the Files panel if order matters for your prompt.

Drag-and-drop is the fastest path for files already open in another app or sitting on your desktop. Useful for screenshots in particular (drag from Preview, drop on Claude) and for quick CSV analysis where the file lives in your Downloads folder right after exporting from a tool. The Add files or photos menu route is better when the files are buried deep in Finder; the drop zone wins when they're already visible somewhere on screen.

Step 3: Manage Attached Files in the Files Panel

Click the file icon in the top-right of the chat to open the Files panel, where every attached file appears as a card with a thumbnail, name, type badge (PDF, CSV, image), and a hover-to-reveal remove control. The panel shows a Content section listing all files currently attached to the chat. PDF cards show a paper preview with a PDF badge; CSV cards show the filename and a CSV badge; image cards show a thumbnail of the image. Click a card to expand it for a closer look at the parsed content or visual preview. The Files panel is the canonical place to remove a wrong file before sending the prompt (hover the card to reveal the X button) or to verify Claude is reading what you think it's reading.

Useful pattern: attach a batch of files first, open the Files panel to confirm each one is correctly parsed (especially CSVs where column inference can go wrong), then type the prompt. Catching a wrong attachment in the Files panel is much cheaper than letting Claude reason on it and having to clarify mid-response. The Files panel persists for the full chat, so files you attached early are still listed when you scroll back days later.

Step 4: Prompt Claude to Summarize a PDF (With a Verification Line)

In the prompt input, ask for a concrete output and include a verification line so you can confirm Claude actually read the file: "Summarize the PDF in five bullet points. Quote one line verbatim from the doc to confirm you read it." The verbatim-quote pattern is the cheapest verification mechanism for PDF analysis: if Claude can produce a matching string from the source, it read the file; if it can't, the summary is hallucinated or based on training-data priors. For longer documents, scope the ask ("Summarize sections 2-4 in five bullet points") so Claude doesn't quietly skip pages to fit your output budget.

Other useful PDF prompts: "Extract every defined term from this contract as a markdown table with term, definition, and the section number where it first appears.", "List every numbered statistic in this report with the page number from the PDF viewer and a one-sentence context note.", "What questions does this paper leave unanswered? List five with the section that raised each one." The pattern is the same: ask for something verifiable, scope it to a part of the document, and reference page numbers using the PDF viewer's count.

Find every Claude conversation by what you said inside it. Claude's native search finds conversations by title. On paid plans (Pro, Max, Team, Enterprise), Claude also offers conversational RAG search via natural-language queries, scoped per-project. AI Toolbox (formerly Claude Toolbox) adds full-text search across all message content with an Exact match toggle on claude.ai, plus message-level bookmarks with scroll-to and highlight, and per-conversation TXT or JSON export. Free plan: 5 search results per query, 2 bookmark conversations. Premium: unlimited at $9.99 per month or $99 lifetime. Working across ChatGPT, Gemini, and Claude? All Access Lifetime ($149) covers all three modules and saves $148 versus three single-platform Lifetimes ($99 x 3 = $297). See the Claude module →

Step 5: Prompt Claude to Analyze a CSV (Ask for a Calculation)

For CSVs, the highest-leverage prompts ask for a calculation across rows and request that Claude show its math: "From the CSV, which keyword has the highest impression-to-click ratio? Show the calculation." CSVs are read column-by-column; Claude infers column types (numeric, categorical, date) on first read and can sort, filter, group, and aggregate them inline. Asking it to show the calculation is the equivalent of the PDF verbatim-quote pattern: you see the work, the column names Claude used, and the formula. If the column names in your CSV are ambiguous (clicks vs impressions vs current_position), being explicit in the prompt about which column to use cuts the back-and-forth.

Other useful CSV prompts: "For each row where ctr is below 0.005, list the keyword, monthly_searches, and current_position in a markdown table sorted by monthly_searches descending.", "Compute the median ctr across the file and flag every row more than two standard deviations below the median.", "Group rows by intent column and report row count plus average ctr per group." The pattern stays consistent: name the columns explicitly, ask for a verifiable output, and request the calculation be shown.

Step 6: Read the Math Claude Shows Its Work

Claude returns a structured response that names the winning row, lists the columns it used, derives the formula it applied, and walks the calculation step by step. In our test, Claude opened a status line ("Derived impression-to-click ratio from CTR data, identified lowest performer"), then surfaced a caveat about the CSV not containing raw impressions or clicks columns (only monthly_searches, current_position, ctr, intent), explained that it treated monthly_searches as the impression proxy and derived clicks = monthly_searches × ctr, and produced the formula impression-to-click ratio = 1 / ctr. It then computed the answer ("how does an ai extension work" with monthly_searches 880, ctr 0.003, clicks 2.64, ratio 333.3) and listed the runners-up. The response ended with an explicit offer to re-run with real impressions data if available.

The structured shape is repeatable: status line, caveats, formula, per-row math, ranked answer, follow-up offer. That shape is what makes Claude useful on CSVs versus a typical chat reply: every claim is auditable in a few seconds. If the math is wrong (wrong column, wrong formula, missing edge case), it's visible immediately. Treat the "show the calculation" prompt as the default for any quantitative CSV question.

Image Uploads: Charts, Screenshots, and Whiteboards

For image uploads, Claude reads charts, diagrams, screenshots, handwritten notes, and product UI directly without OCR as a separate step. Useful prompt patterns: "Describe everything in this screenshot, including labels, colors, and layout.", "Transcribe the handwritten notes in this image.", "This chart shows quarterly revenue. Extract every datapoint into a markdown table with columns Quarter, Revenue, YoY change." Image quality matters more than for text uploads; aim for 1000x1000 pixels minimum per Anthropic's guidance. Compressed or thumbnail-sized images degrade accuracy especially for small text or fine chart labels.

One practical use case: paste a screenshot of an analytics dashboard, ask Claude to translate the visible metrics into a markdown table, then ask follow-up questions over the table inline. Cheaper than CSV exports, faster than typing the numbers out, and Claude tolerates the dashboard's visual chrome (logos, sidebar, headers) without needing you to crop the image first.

Claude File Uploads vs ChatGPT and Gemini in 2026

Claude's file upload is the most generous of the three big assistants in 2026: 500MB per file vs ChatGPT's 512MB per file and Gemini's 100MB per file for most types. Where Claude wins is the multimodal PDF mode under 100 pages (visual elements like charts and graphics get read directly, not just OCR'd), the 20-files-per-chat limit (ChatGPT caps at 10, Gemini at 10), and the explicit Project-files lane for unlimited reference material that fits the context window. Where ChatGPT wins is the file persistence across new chats inside a Project and the Advanced Data Analysis sandbox that runs actual Python on uploaded files. Where Gemini wins is the Google Drive native integration and the deep image-generation pipeline tied to upload context.

Frequently Asked Questions

How do I upload a file to Claude?

Click the plus icon next to the chat input and choose Add files or photos, or drag the file directly onto the chat window. Claude accepts PDFs, CSV, TXT, DOCX, XLSX, PPTX, and common image formats up to 500MB per file and 20 files per chat. The upload works the same on claude.ai in the browser and inside the Claude Desktop app on macOS and Windows.

What file types does Claude support in 2026?

Claude accepts PDF, CSV, TXT, DOCX, XLSX, PPTX, and common image formats (JPG, PNG, GIF, WEBP). PDFs under 100 pages run in multimodal mode (Claude reads both text and visual elements); PDFs over 1000 pages drop to text-only. Images cap at 8000x8000 pixels with at least 1000x1000 pixels recommended for best results.

What is the maximum file size on Claude?

500MB per file for chat uploads, with 20 files per chat as the count cap. Project files are stricter at 30MB per file but the count is unlimited as long as the total content fits Claude's context window. Anthropic's official guidance: for larger documents, consider dividing them into smaller sections to stay within limits.

Can Claude read images and screenshots?

Yes. Claude reads charts, diagrams, screenshots, handwritten notes, and product UI directly without a separate OCR step. Image dimensions cap at 8000x8000 pixels; Anthropic recommends at least 1000x1000 pixels because small or low-resolution images degrade analysis quality. Compressed thumbnails and low-DPI captures will lose accuracy especially on small text or fine chart labels.

Can Claude analyze CSVs and spreadsheets?

Yes. Claude reads CSV and XLSX files column-by-column, infers types on first read, and can sort, filter, group, and aggregate inline. Ask Claude to show the calculation when requesting a numerical answer so you can verify the column choice and formula. Naming the columns explicitly in the prompt cuts ambiguity on files with multiple similarly-named columns.

Does Claude read every page of a long PDF?

Claude reads up to 100 pages in multimodal mode (text plus visual elements). Between 100 and 1000 pages, multimodal is best-effort and Claude may quietly skip pages to fit your prompt's output budget. Over 1000 pages, Claude drops to text-only extraction. Scoping your prompt to specific sections ("Summarize sections 2-4...") helps Claude allocate its attention deliberately.

How are Project files different from chat uploads?

Project files attach to a Claude Project and persist across every chat inside that project, with 30MB per file but unlimited count as long as the total fits the context window. Chat uploads attach to a single chat at 500MB per file, 20 files. Project files are text extraction only except for multimodal PDFs; for image visual analysis, attach to a chat directly.

Can I drag and drop files into Claude?

Yes. Drag any file from Finder, your Desktop, or any other window onto the Claude chat window. A "Drop files here to add to chat" overlay appears in the center of the window; release to attach. Multi-file drags work too: select multiple files in Finder, drag the group onto Claude, drop. Drag-and-drop accepts the same file types and limits as the plus menu route.

What Claude Does Not Do (and Where AI Toolbox Picks Up)

Claude's file upload is generous on size and file count, but the platform still does not let you search across every chat by message content (native search finds chats by title), bookmark individual messages inside a long file-analysis reply for later review, or export a single conversation to TXT or JSON from the chat UI. The Claude module of AI Toolbox (formerly Claude Toolbox) adds all three on claude.ai in any Chromium browser: a full-text search popup that scans every synced message with an Exact match toggle, message-level bookmarks (2 conversations free, up to 1,000 bookmarks on Premium) with click-to-scroll and yellow highlight, and per-conversation TXT or JSON export.

Pricing matches the rest of AI Toolbox: free (5 search results per query, 2 bookmark conversations) or Premium at $9.99 per month or $99 lifetime. If you also use ChatGPT or Gemini, All Access Lifetime ($149) covers the ChatGPT, Gemini, and Claude modules in one install and saves $148 versus three separate Lifetimes ($99 x 3 = $297). AI Toolbox is Chromium-only (Chrome, Edge, Brave, Opera, Arc) and works on claude.ai. Pair it with the Claude Desktop app and Computer Use Beta for global hotkey and agent-style automation on top of the file upload flow.

Install AI Toolbox for Claude from the Chrome Web Store to get full-text search, message bookmarks, and TXT or JSON export on claude.ai. The same install also adds folders, pins, prompts, and prompt chaining on chatgpt.com, plus full-text search and export on gemini.google.com.

Last updated: June 1, 2026