modelcontextprotocol/ext-apps

Overview

This skill lets the agent open, navigate, and interact with PDF documents via a local PDF server. It supports listing available PDFs, opening a specific file or URL, and starting on a given page. Use it as the primary tool whenever the user wants to read, search, or analyze a paper or document.

How this skill works

The skill exposes two core actions: list_pdfs to show available documents and display_pdf to open a PDF in an interactive viewer. If a user provides a URL or local path, display_pdf is called directly (optionally with a starting page). If the user asks to open a document without specifying which, the skill calls list_pdfs so the user can pick.

When to use it

• User asks to open or read a specific PDF, paper, or report (provides URL or path).
• User requests to browse available local PDFs before selecting one to read.
• User asks to read a particular page, section, or start point in a PDF.
• User wants the agent to summarize, extract data, or answer questions about a document.
• User requests to open an arXiv/bioRxiv/medRxiv/Zenodo/other supported source by identifier.

Best practices

• Always display the PDF first so the agent and user reference the same document.
• For long or multi-section documents, ask which page range or section the user cares about.
• When given an arXiv ID or supported repository ID, convert it to the PDF URL automatically.
• Compare papers one at a time in the viewer and call out specific pages or figures when noting differences.
• Confirm file selection when list_pdfs returns multiple similarly named documents.

Example use cases

• User says 'open arxiv 2301.12345' — convert to PDF URL and call display_pdf starting at page 1.
• User asks 'open the latest report' with no path — call list_pdfs so the user can choose a file.
• User requests 'summarize pages 10–15' after a PDF is displayed — navigate to those pages and extract key points.
• User wants data extraction from a methods section — ask which section or page, then display and extract.
• User asks to compare two papers — display the first, note key findings, then display the second and highlight differences.