VaultPDF MCP 🔒

Local PDF tools for AI agents. An MCP server that lets Claude, ChatGPT Apps, or any MCP host work with PDFs on the user's own machine — files are never uploaded.

Part of the VaultPDF family of privacy-first tools. Built early for the fast-growing, still-uncrowded MCP / agent-app ecosystem.

Tools

tool	what it does
pdf_info	page count + per-page dimensions of a PDF
pdf_merge	merge several PDFs into one, in order
pdf_split	extract pages (e.g. 1-3,5,8) into a new PDF
images_to_pdf	combine JPG/PNG images into a PDF, one per page

All operations run locally via pdf-lib. Nothing leaves the machine.

Run

# from npm (after publish):
npx vaultpdf-mcp
# or straight from GitHub, no install:
npx github:AlexeySamosadov/vaultpdf-mcp

Use in Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "vaultpdf": { "command": "npx", "args": ["github:AlexeySamosadov/vaultpdf-mcp"] }
  }
}

Then ask Claude: "merge /path/a.pdf and /path/b.pdf into /path/out.pdf".

Distribution (early = low competition)

• MCP Registry + awesome-mcp-servers (PRs).
• npm as vaultpdf-mcp (so npx vaultpdf-mcp works) — needs an npm login.
• ChatGPT App via the Apps SDK (wrap this MCP server behind a hosted endpoint) — needs an OpenAI developer account.
• Claude Skills directory entry. Same capability, listed everywhere — that's how agent tools get reach in 2026.

Why

PDF manipulation is one of the most common things people ask AI agents to do, and almost no MCP server does it locally / privately. Free + open to drive adoption; funnels to the paid VaultPDF web tools and a future hosted Pro tier.

MIT licensed.