How to Work Around Claude's File Upload Limits

Claude accepts files of up to 500MB each in a chat, with up to 20 files per chat. Files you add to a Project have a lower cap of 30MB each. Full breakdown by where the file goes:

• Chat uploads: 500MB per file, up to 20 files per chat, images up to 8,000 x 8,000 pixels, PDFs up to 1,000 pages
• Project files: 30MB per file; no fixed number of files, but the total content must fit within Claude's context window; text extraction only, except for multimodal PDFs
• Files API (developers): 500MB per file and 1 TB of storage per organization, stored server-side and referenced by ID
• Images: JPEG, PNG, GIF, WebP up to 8,000 x 8,000 pixels
• PDFs: 100 pages or fewer for full visual analysis (text + charts + images); 101 to 1,000 pages are read as text only; over 1,000 pages are rejected

Anthropic's help center gives one set of upload limits, with no difference between models. What changes by model is the context window, covered below. If you'd rather keep large files and document sets in one place that Claude can reach from any session, see Storage for AI Agents: a Fastio workspace stores single files of up to 25, 50 or 100 GB depending on plan. That doesn't raise Claude's own limits. The files stay in the workspace, Claude pulls the passages it needs through MCP, and whatever it reads still has to fit in its context window. The Claude-specific setup guide walks through connecting Claude to a workspace.

Keep your files where Claude can reach them

Fastio keeps your files in a workspace that Claude reaches over MCP, with no re-uploading between sessions and built-in RAG indexing. Claude's own upload limits still apply to files you attach in a chat.

Start 14-Day Trial

• Copilot File Upload Limits, Formats, and Workarounds for Large FilesThe Copilot file upload limit restricts direct document attachments to 10 MB per file across standard chat interfaces,...
• Claude Code File Size Limits: CLI Truncation and Large File HandlingClaude Code imposes strict operational file size limits in the terminal, truncating file reads and command outputs past...
• Managing Claude Code Daily Limits: Spend Caps, Quotas, and Unattended WorkflowsAutonomous coding agents can rapidly exhaust token quotas and project context when executing unbounded multi-turn loops...
• Claude Code Rate Limits: 5-Hour Usage Caps, 429 Errors, and WorkaroundsClaude Code rate limits enforce execution thresholds across terminal requests, tokens per minute, and five-hour rolling...
• How to Handle Claude Code Token Limits in Large ProjectsClaude Code enforces a 200,000-token context window that fills quickly on complex repositories as file reads and bash...
• How to Connect Claude to Google Drive: A Complete GuideGetting your Google Drive files into Claude's context is more annoying than it should be. This guide covers the main...

More on this subject: Claude and Claude Code (280 guides)

Claude accepts 10 document formats and 4 image formats. Here's the full list to save you a failed upload.

Documents:

• PDF (.pdf)
• Microsoft Word (.docx)
• Excel spreadsheets (.xlsx), once code execution and file creation are turned on in your account
• CSV (.csv)
• Plain text (.txt)
• HTML (.html)
• OpenDocument Text (.odt)
• Rich Text Format (.rtf)
• EPUB (.epub)
• JSON (.json)

Images:

• JPEG (.jpg, .jpeg)
• PNG (.png)
• GIF (.gif)
• WebP (.webp)

Claude extracts only the text from documents other than PDFs, so images embedded in a Word file or an EPUB won't be read. Anthropic's list doesn't name Markdown or source-code files (.py, .js, .ts, .css, and similar); if one of those is rejected, saving a copy as .txt gets it in as plain text. One practical tip: if you have an Excel file with multiple sheets, converting it to CSV before uploading usually gives Claude better results. CSV strips out formatting overhead and reduces token consumption, so you get more useful analysis per file.

File size is only the first gate. Even when every file is under the cap, you can still hit limits because of how Claude processes text internally. Claude converts every uploaded file into tokens, and Anthropic notes that additional token limits may apply based on the length of the extracted content. A dense 5MB PDF full of text can generate more tokens than a 20MB PDF packed with images.

Those tokens share the context window with your conversation history and Claude's responses. On paid plans, Anthropic lists context windows from 200,000 tokens up to 1 million tokens, depending on the model you chat with. With code execution turned on, Claude summarizes earlier messages as a conversation nears the limit so it can keep going, and the full history stays available for reference. Without code execution, or with a very large first message, you can still run into the limit outright.

Either way, the window is shared, which makes context, not file size, the real bottleneck for anyone working with large document sets. A handful of long, text-heavy PDFs can add up to more tokens than the window holds, even when each one is well under 500MB.

Projects let you build a persistent knowledge base, and they're available on every Claude plan, including free accounts (free users can create up to five). Files stay available across multiple conversations without re-uploading. Project files are capped at 30MB each, well below the 500MB chat limit, but there's no fixed number of files: the total content just has to fit within Claude's context window. On paid plans, Projects switch to retrieval (RAG) as project knowledge approaches that limit, which Anthropic says expands capacity by up to 10x. In that mode Claude pulls only the relevant sections from Project files into its context window at query time, so it can't read the entire knowledge base at once.

Projects work well for reference documentation you query repeatedly, code repositories with many small files, and research papers you want to cross-reference. They don't help with files larger than 30MB, datasets that need to be analyzed as a whole, or binary files like video and audio.

When a file is over the limit for where it's going (500MB in a chat, 30MB in a Project, or more than 1,000 pages for a PDF), the simplest fix is to break it into smaller pieces before uploading.

PDFs: Use Smallpdf or Adobe Acrobat to split by page range. Keeping each part to 100 pages or fewer also means Claude analyzes its charts and images, not just its text. Upload the sections across separate messages or add them to a Project.

CSV and JSON: Split by row count. A 50MB CSV with 500,000 rows becomes two 25MB files of 250,000 rows each, small enough to add to a Project. Python's pandas library handles this in a few lines:

import pandas as pd
df = pd.read_csv("large_file.csv")
df.iloc[:250000].to_csv("part1.csv", index=False)
df.iloc[250000:].to_csv("part2.csv", index=False)

Images: Compress before uploading. Claude's vision capabilities work fine at 1080p, and Anthropic recommends images of at least 1,000 x 1,000 pixels, so don't shrink below that. Downscaling a 4K image from 15MB to 2MB rarely affects analysis quality. This approach works for occasional large files, but gets tedious if you're regularly dealing with document bundles above the limit.

For a more permanent solution, keep your files in external cloud storage and let Claude access them through the Model Context Protocol (MCP). Instead of uploading files into the chat, you store them on a remote platform and give Claude direct read access via MCP tools. The model pulls only the text it needs at query time. Claude's upload limits still apply to anything you attach to a chat, and whatever Claude reads still has to fit in its context window, but the files themselves stay in storage, so you stop re-uploading them for every conversation.

Cloud storage options that work for this pattern:

• AWS S3 + MCP, Flexible, durable, widely supported. Requires AWS account and credentials setup. Multiple MCP servers exist for S3 access.
• Google Drive, Easy to set up. Claude can access Drive files via MCP with the right server configured.
• Fastio, Storage built specifically for AI agent workflows. A consolidated MCP toolset, Intelligence Mode for automatic RAG indexing.

How to set up Fastio for this:

1. Start a 14-day Business Trial on Fastio
2. Upload your files to a workspace and enable Intelligence Mode for automatic RAG indexing
3. Connect Claude to the Fastio MCP server (supports Streamable HTTP and SSE)
4. Ask Claude questions about your files, the MCP integration handles retrieval, and Intelligence Mode returns answers with source citations

Any MCP-compatible model can query the same files, so the same setup works with ChatGPT, Gemini, or other clients that support MCP.

Use Claude Projects if you have many files under 30MB each that you query repeatedly. Projects come with every Claude plan, and paid plans add retrieval for larger knowledge bases. No extra setup required.

Use file splitting if you occasionally need to analyze a single file that's over the limit and can be logically divided, such as a multi-chapter PDF or a big CSV, and you'd rather not set up additional infrastructure.

Use external storage with MCP if you regularly work with large files, need a persistent knowledge base across sessions, want the same files to work with multiple AI models, or are building agents that need programmatic file access.

For most developers doing serious document work, external storage ends up being the right answer. It ends the re-upload cycle and adds semantic search across a whole workspace.