Claude File Size Limits: Chat, Projects, and Context Ceilings Explained

Claude enforces an upload ceiling of 500MB per file across standard chats and 30MB per file within Claude Projects, according to official Anthropic documentation updated in September 2026.

The Claude file size limit is 500MB per file in standard chats (up to 20 files) and 30MB per file in Claude Projects, with overall document capacity constrained by Claude's 200,000-token context window.

Understanding these boundaries is essential for technical leads, developers, and knowledge workers who rely on Anthropic's models for document analysis. While the front-end interface accepts individual attachments through a simple file picker, file size caps, page count ceilings, and context window mechanics govern what the model can ingest.

A frequent source of confusion comes from contrasting standard conversation threads against Claude Projects. In standard web and desktop chat conversations, Claude allows users to attach up to 20 files per chat, with each file supporting a maximum size of 500MB. This 500MB Claude chat file upload limit allows users to upload extensive technical manuals, raw source code packages, research papers, or multimodal datasets directly into a single conversation.

In Claude Projects, the operating rules shift. Claude Projects serves as a persistent workspace where Pro, Team, and Enterprise users assemble knowledge bases shared across multiple conversations. Anthropic restricts individual project files to 30MB per file.

Many competitor articles and third-party guides claim that Claude Projects imposes a fixed file count cap, such as 10 or 25 documents. That claim is incorrect. Anthropic enforces no numerical cap on the total count of files uploaded to a project. A project can hold 5 files, 50 files, or hundreds of files, provided two boundaries are respected: each individual file must remain under the 30MB ceiling, and the cumulative text across all project files must fit within Claude's 200,000-token context window.

The following comparison table outlines documented limits across Claude's user interfaces and developer endpoints:

When uploading files through the web interface, users can click the plus icon in the bottom left corner of the chat input box, select Add files or photos, and choose assets from local storage. Alternatively, dragging and dropping files directly onto the browser window or pasting image data from the clipboard initiates an upload. Files uploaded to a standard chat remain local to that specific conversation thread, whereas files added to a project's Files section persist across all chats created within that project.

• 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...
• Cursor Composer Limits: Request Caps, File Context, and Scaling SolutionsCursor Composer limit refers to the combined threshold of fast request quotas, concurrent file edits, and active...
• Claude Code Memory Limits: CLAUDE.md Sizing, Auto-Memory, and Context WorkaroundsClaude Code memory limits represent the size thresholds and retention boundaries governing CLAUDE.md files, persistent...
• 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...
• Claude Attachment Limit: File Caps, Size Ceilings, and Knowledge RetrievalThe Claude attachment limit restricts web chat uploads to 20 files per conversation with a 500MB maximum file size,...
• 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...

More on this subject: Claude and Claude Code (255 guides)

While file size represents the primary gate during upload, file type and internal document structure determine whether Claude successfully parses your content. Anthropic supports a wide variety of formats, but processing rules vary depending on whether an asset is a structured table, an image, a PDF, or a text document.

Supported Document and Image Formats

Claude natively processes the following document formats:

• PDF Documents (.pdf): Evaluated for both text and visual layouts depending on document length.
• Word Documents (.docx): Ingested through plain text extraction; formatting styles and layout metadata are discarded.
• Structured Data (.csv and .json): Parsed as tabular text or structured key-value objects for code analysis and data manipulation.
• Plain and Rich Text (.txt, .rtf, .odt, and .html): Stripped of markup and ingested as clean textual tokens.
• Digital Books (.epub): Read through linear text extraction for long-form reference.
• Spreadsheets (.xlsx): Supported when code execution and file creation are enabled in account settings.

For image files, Claude supports JPEG, PNG, static GIF, and WebP formats. Uploaded images support dimensions reaching 8,000 by 8,000 pixels. Animated GIFs are accepted, but Claude evaluates only the first frame and discards subsequent motion frames. Vector graphics (SVG), uncompressed bitmaps (BMP, TIFF), and camera RAW formats cannot be parsed directly and must be converted to PNG or JPEG before uploading.

PDF Processing Rules and the 1,000-Page Boundary

PDF documents receive unique handling within Anthropic's ingestion pipeline. Claude alters its processing behavior based on document page count:

• PDFs of 100 Pages or Fewer: Claude processes both text and visual components. The model inspects embedded charts, graphs, flowcharts, photographs, and table formatting alongside surrounding prose. This allows users to ask questions about visual diagrams, architectural schematics, and formatted financial reports.
• PDFs from 101 to 1,000 Pages: Claude shifts exclusively to text extraction. Embedded charts, illustrations, diagrams, and graphic elements are discarded during ingestion. The model reads the raw text stream, allowing it to summarize long books, transcripts, and technical whitepapers while bypassing visual token costs.
• PDFs Exceeding 1,000 Pages: Claude rejects the file immediately. The front-end upload dialog aborts and returns an explicit error stating that the uploaded file is too large.

Common Upload Errors and How to Resolve Them

When an upload fails, Claude displays concise error banners that point to specific transport or parsing violations:

1. "Uploaded file is too large": This error occurs when an individual file exceeds 500MB in a standard chat, when a file exceeds 30MB in Claude Projects, or when a PDF document exceeds 1,000 pages. To fix this error, compress high-resolution images, remove unused attachments, or split long PDF files into sequential parts before uploading.
2. "Too many files": Standard chat threads accept a maximum of 20 files per conversation. If you drag 25 files into the composer, Claude blocks the submission. To proceed, group related text files into a single consolidated file or move your workflow into Claude Projects.
3. Spreadsheet Ingestion Blocked: If you upload an XLSX file and encounter a failure, check your account settings. Spreadsheets require Claude's code execution tool to parse workbook sheets, calculate formulas, and run data analyses. Enabling code execution in your profile resolves the issue.
4. Non-PDF Embedded Image Loss: Claude extracts text only from DOCX, RTF, and HTML files. If your Word document relies on screenshots or diagrams embedded between paragraphs, Claude will not see them. Export the document as a PDF within the 100-page threshold to ensure visual elements reach the vision engine.

Passing the 500MB or 30MB file size boundary does not guarantee that Claude can reason over your entire document. The true ceiling on Claude's comprehension is not file size on disk, but token capacity in the model context window.

File Size versus Token Footprint

File size measures the physical bytes stored on disk, whereas tokens represent the semantic fragments processed by a transformer model. A compact plain text file or dense source code archive can easily contain more tokens than Claude can hold, while a large PDF containing high-resolution scanned graphics might yield only a modest amount of extracted text once visual downsampling finishes.

Code repositories and tabular data files like CSV or JSON are dense in token footprint. Every comma, bracket, indentation space, and numeric string consumes tokens rapidly. In contrast, plain prose benefits from standard subword tokenization, but long narrative archives still accumulate substantial token weights.

Claude models, including Claude 3.5 Sonnet and Claude 3.7 Sonnet, operate with a standard context window of 200,000 tokens. This 200,000-token envelope is a shared allocation: it must accommodate system instructions, user prompts, conversation history, attached documents, project knowledge files, and the model's generated response.

The Dynamics of Project Knowledge Saturation

In Claude Projects, all uploaded files are packed into the prompt payload on every conversational turn. While Anthropic allows an unlimited count of files up to 30MB each, the cumulative text extracted from those files eats into the 200,000-token budget.

When project files consume a large portion of the context budget:

• Every interaction in that project starts with most of its context window already filled.
• The model has limited room remaining for user prompts, multi-turn chat history, tool use, and generated responses.
• Saturated context increases processing latency and can degrade retrieval precision, a phenomenon known as in-context attention distraction.
• If user prompts and conversation history push the total volume past 200,000 tokens, Claude truncates earlier messages or refuses to complete the generation.

Relying on direct file uploads for large documentation libraries, multi-year financial archives, or large codebases eventually collides with this token ceiling. Splitting files into smaller chunks provides temporary relief, but it fragments cross-document reasoning and requires tedious manual organization.

Index Enterprise Corpora Beyond Claude Upload Caps

Connect your team files to Claude using an intelligent Fast.io workspace. Search large document collections via MCP without hitting chat upload limits or bloating context windows. Every organization starts with a 14-day free trial, which requires a credit card.

Start 14-Day Trial

When document collections exceed 30MB per file, 500MB per chat, or 200,000 tokens in aggregate, teams must separate persistent storage from prompt context. Instead of forcing entire file contents into Claude's conversational payload, modern engineering architectures position an intelligent retrieval substrate between the files and the model.

Traditional Workarounds and Their Limitations

Before implementing unified workspace retrieval, teams typically test several intermediate workarounds:

• Manual File Splitting Scripts: Developers often write local Python scripts using libraries like PyPDF or pdfplumber to divide 2,000-page manuals into 50-page segments. While this satisfies the 1,000-page limit, it creates hundreds of disconnected files. Cross-document synthesis becomes impossible because Claude cannot search across fragments that sit unattached on local disk.
• Custom Vector Databases: Engineering teams frequently build dedicated retrieval pipelines using Pinecone, Weaviate, or pgvector. While effective, custom vector databases require managing embedding models, tuning chunk boundaries, maintaining syncing pipelines, and paying dedicated infrastructure hosting costs.
• Consumer Cloud Drive Connectors: Syncing Google Drive, Dropbox, or Box folders directly into chat interfaces often recreates the original context problem. Connectors that dump entire files into prompt context rapidly consume token budgets without delivering targeted answers.

Connecting Claude to an Intelligent Fast.io Workspace via MCP

Fast.io provides an intelligent workspace platform designed for agentic teams and knowledge workers. Fast.io leaves Anthropic's native upload limits exactly where they are; what it adds is an indexed, searchable home for the documents that do not fit into prompt memory.

The architecture operates through three unified components:

1. Workspace Storage and Cloud Sync: Teams store enterprise files in shared, organization-owned Fast.io workspaces. Files can be uploaded directly or imported from existing repositories. Fast.io supports Cloud Sync for Google Drive, Dropbox, Box, and OneDrive on a schedule or on demand. Teams can also use URL import to ingest assets from external web endpoints without consuming local bandwidth.
2. Intelligence Mode and Hybrid Search: When you enable Intelligence Mode on a Fast.io workspace, the system automatically indexes all documents, spreadsheets, and manuals upon arrival. Fast.io runs hybrid search combining full-text search, semantic search, and metadata value filtering. No vector database configuration or manual chunking is required.
3. Model Context Protocol (MCP) Integration: Claude connects to Fast.io through the remote MCP server at https://mcp.fast.io/mcp/tools over Streamable HTTP. Claude (web, desktop, mobile): open Customize, then Connectors; add a custom connector and paste https://mcp.fast.io/mcp/tools; select Connect and sign in to Fastio in the window that opens; in a chat, turn Fastio on from the + menu under Connectors. Sign-in shows a Review Permissions screen where the person picks Read Only or Read & Write and which organizations and workspaces the connection can reach. For setup steps, see the Fast.io documentation. The Fast.io MCP server provides a storage tool.

When Claude needs information from an extensive regulatory filing or an archive of architectural blueprints, it calls the MCP storage tool using the search action:

{
  "name": "storage",
  "arguments": {
    "action": "search",
    "query": "annual maintenance schedule requirements for industrial chillers",
    "workspace_id": "ws_enterprise_archive"
  }
}

Instead of ingesting thousands of pages of text, Claude receives only the specific paragraphs, table data, and file references relevant to the prompt. Queries execute with targeted precision, preserving context capacity and eliminating upload errors.

Structured Extraction with Metadata Views

For document collections like contracts, insurance claims, and invoices, unstructured search is only part of the solution. Fast.io offers Metadata Views, which turn messy document folders into live, queryable databases.

Users describe the fields they want extracted in natural language, such as contract effective dates, governing laws, counterparties, or payment totals. Fast.io parses the documents and populates a typed, filterable table supporting text, integer, decimal, boolean, URL, JSON, and timestamp schemas. AI assistants can query these structured metadata views directly via MCP, retrieving exact figures without reading raw documents.

Moving beyond raw file uploads transforms how teams govern sensitive business records. Individual chat threads isolate knowledge on a single user machine, creating data silos and security risks. Managing enterprise documents in shared Fast.io workspaces provides team governance, auditability, and access control.

Per-File Version Control and Audit Logging

In collaborative environments where team members and autonomous agents collaborate, tracking changes is critical:

• Per-File Version History: Every update to a document creates a new version while preserving complete prior states. If an agent refactors a document or a user modifies a specification, prior versions can be inspected, compared, and restored at any time.
• Detailed activity log: Fast.io records every workspace interaction in a detailed activity log. Administrators can review who accessed, uploaded, modified, downloaded, or shared any file across the organization.

Granular Access Control and Branded Client Sharing

Enterprise data security requires strict boundaries between internal project teams, client stakeholders, and automated AI agents:

• Tiered Permissions: Administrators configure access rights at the organization, workspace, folder, or file level. An AI agent can be granted read-only access to reference folders while maintaining write permissions in an output directory.
• Branded Shares: When sharing assets with external clients, teams generate branded Send, Receive, or Exchange shares. Shares can be configured with expiration dates, per-recipient access controls, and disabled file downloads to safeguard proprietary research.
• Collaborative Notes: Humans and AI agents can co-author project documentation, summaries, and meeting briefs inside real-time Collaborative Notes.
• Activity Feeds: External systems and automated workers can monitor workspace activity in real time using WebSocket feeds or HTTP activity long-polling (GET /current/activity/poll/{entity_id}?wait=95&lastactivity={timestamp}), reacting instantly as new files arrive.
• E-Signature Workflows: When legal agreements or project sign-offs are required, teams can send documents for signature directly from the workspace through Fast.io E-Signature. The platform supports sequential or parallel signers, optional one-time passcodes, a tamper-evident audit certificate, and automatic filing of executed contracts back into the workspace.

Subscription Plans and Starting with a Free Trial

Every organization starts with a 14-day free trial, which requires a credit card. Paid subscriptions are structured across Starter, Business, and Enterprise plans. Credits meter AI work. Storage and team seats come with each subscription plan. Learn more about agent storage architecture at Fast.io Storage for Agents and explore tier details on the Fast.io Pricing page.

By pairing Claude's analytical strengths with Fast.io's persistent workspace infrastructure, teams overcome conversational upload boundaries and build durable knowledge foundations for their AI workflows.