PDF Tool (Anthropic Computer Use)

What it does

The PDF skill enables Claude computer use agents to interact with PDF documents. When invoked, Claude opens the PDF in a browser tab or desktop viewer, scrolls through pages, extracts text, and can answer questions or produce summaries.

This is Anthropic’s reference implementation from the computer-use-demo quickstart, adapted as a standalone SKILL.md for direct use in Claude Code and compatible agents.

Installation

npx cache add anthropic-pdf
# or clone directly:
# git clone https://github.com/anthropics/anthropic-quickstarts

Triggers

The skill activates when Claude detects phrases related to PDF reading:

• “read this pdf”
• “extract text from [filename]”
• “open the pdf at [path]”
• “summarize this document”

Requirements

• Claude computer use beta access
• A display environment (Xvfb or real desktop for desktop mode, or a browser for web mode)
• The PDF must be accessible via a local path or URL

Behavior

1. Claude opens the file in the browser using open -a "Google Chrome" [path] (macOS) or equivalent
2. Scrolls through each page, reading visible text
3. Constructs a running transcript
4. Returns extracted text or answers the user’s query

Compatibility

Works natively with Claude claude-code agents via the computer use API. Cline users can invoke it with partial functionality — the file opening works but page-by-page scroll may require adaptation.

Cursor and Aider do not support the computer use tool suite and cannot run this skill.