Filevolo for AI assistants

A small program that gives Claude, or any assistant that speaks the Model Context Protocol (MCP), eleven PDF tools. Ask it to "merge these three contracts and drop the cover page" and it does it on your computer.

Download filevolo-mcp.mjsOne file, about 5 MB. Needs Node.js 20 or newer.

It runs on your computer, not ours

The assistant starts the program on your machine and talks to it locally. Your PDFs are read from your disk and the results are written to your disk; nothing is uploaded to Filevolo or anywhere else by this program. That is why there is no online API: a service that processed your files on our servers would break the promise the rest of Filevolo makes. The assistant itself may still send what it reads to its own provider, as with any file you share with it, so use it on documents you are comfortable sharing with that assistant.

Set it up

Save the file somewhere permanent, then add it to your assistant. For Claude Desktop, open Settings, Developer, Edit Config, and add:

{
  "mcpServers": {
    "filevolo": {
      "command": "node",
      "args": ["C:\\Users\\you\\filevolo-mcp.mjs"]
    }
  }
}

Use the real path to the file (on a Mac or Linux, something like /Users/you/filevolo-mcp.mjs), then restart the app. For Claude Code:

claude mcp add filevolo -- node /path/to/filevolo-mcp.mjs

Other assistants that support local (stdio) MCP servers take the same command: node with the path as its argument.

What it can do

pdf_info: Page count, page size, title, author, and whether the PDF has selectable text.
pdf_extract_text: The text of a PDF as clean paragraphs, optionally saved to a .txt file.
pdf_summarize: The most representative sentences of a PDF, word for word.
pdf_merge: Join several PDFs into one.
pdf_extract_pages: Keep only chosen pages, as a new PDF.
pdf_delete_pages: Save a copy without chosen pages.
pdf_split: Split into several files by page ranges.
pdf_rotate: Turn pages by 90, 180 or 270 degrees.
pdf_crop: Trim margins from pages.
pdf_remove_metadata: Clear title, author and other document properties.
pdf_to_word: Convert a PDF with selectable text into an editable Word file.

Safe by design

It never overwrites a file. Results are saved next to the original with a suffix, numbered if the name is taken.
An existing output path is refused unless the assistant is told, explicitly, to replace it, and an input file can never be replaced.
It only reads and writes files at paths the assistant gives it, and only PDFs. It does not browse your disk, run commands or open network connections.
Password-protected PDFs are reported, not cracked. A scan has no text to read and needs OCR first.

The same tools are available in the browser on the PDF tools page, with a few more (editing, signing, OCR, translation).