PDF

PDF is the document format people use for papers, reports, contracts and forms. The official @modelcontextprotocol server opens an interactive PDF viewer and supports local files plus remote URLs from approved academic sources such as arXiv, bioRxiv, Zenodo and others.

What it does: your AI can list available PDFs, display a document, search inside it, move to a page, extract text from selected pages, take a screenshot, add highlights or notes, stamp pages, fill forms and save an annotated copy when the file is in an allowed local root. The README describes the core pattern as "chunked pagination" so large PDFs do not need to travel in one oversized tool call.

What you can ask

• "Open this research paper and summarize the argument from pages 1 to 3."
• "Search this contract for cancellation terms and highlight the important paragraph."
• "Add a confidential stamp to every page and save a new annotated copy."

What configuration you need

You do not need a token. The connector runs locally with npx and can only read local files you pass to it, local folders exposed by client roots, or remote PDFs from the allowed source list.

1. Install it from the Lab like any connector without secrets.
2. Open PDFs only from folders you are comfortable exposing to the agent.
3. For saving annotations, use a mounted local directory root; the server refuses writes outside allowed roots.

Be careful with confidential documents. This connector makes PDF contents available to the AI session, so only open files that belong in that workspace.