QuillPDF

For AI agents & developers

Give your AI agent PDF hands — locally

quillpdf-mcpis an open-source (MIT) Model Context Protocol server and CLI: merge, split, rotate, watermark, Bates-number, clean metadata, page-count. Every operation reads and writes PDFs on your own disk. stdio transport only — it never opens a network connection, so your agent's documents never leave your machine.

View on GitHubOur claims discipline

Why a local MCP server

Agents increasingly do document work, and most PDF APIs answer with "upload it to us." For legal discovery, medical records, or financials, that's a non-starter. quillpdf-mcp inverts it: the operation runs where the file already is. Your assistant calls a tool, the PDF gets merged or stamped or scrubbed on your disk, and nothing transits anyone's server — ours included.

Seven operations, two interfaces

Merge · split · rotate · watermark · Bates-number · clean metadata · page-count. Each is available two ways:

  • MCP server— register it with any MCP-capable assistant (Claude Desktop, Claude Code, and others) and the tools appear in the agent's toolkit.
  • CLI — the same operations as a plain command-line tool for scripts and pipelines.

Install

From source (Node.js 18+):

git clone https://github.com/PurpleDirective/quillpdf-mcp
cd quillpdf-mcp
npm install && npm run build
node dist/cli.js --help

Full MCP client configuration examples — including the JSON snippet for Claude Desktop — live in the README.

Trust properties you can check

  • MIT licensed, full source on GitHub — read what it does.
  • stdio transport only; no network I/O anywhere in the codebase.
  • Small dependency surface built on pdf-lib — auditable in an afternoon.
  • Same claims discipline as the website's transparency page: what it doesn't do yet (redact, compress, OCR — phase 2) is stated, not implied away.

Questions developers ask

What is quillpdf-mcp?

An open-source (MIT) Model Context Protocol server and command-line tool that gives AI agents local PDF operations: merge, split, rotate, watermark, Bates numbering, metadata cleaning, and page-count. It runs on your machine, reads and writes files on your disk, and uses stdio transport only — it never opens a network connection or transmits your documents.

Why does local matter for agent PDF tooling?

Most PDF APIs require uploading documents to someone else's server. If an agent is handling legal discovery, medical records, or financials, that upload is often unacceptable. quillpdf-mcp runs the operation where the file already is — the bytes stay on your disk, and the agent gets the result.

How is this related to the quillpdf.com website?

Same team, same principle, different runtime. The website runs PDF tools in your browser tab; quillpdf-mcp runs them in your local environment where an AI assistant or script can call them. Neither uploads your files anywhere.

What's on the roadmap?

Redaction, compression, and OCR are the planned phase-2 operations — the same capabilities the browser tools already have. The repo's issues page is the honest source of truth for status.