PaperPilot

What it does

• MCP stdio server (paperpilot-mcp) with ready-made client configs: `paperpilot-mcp --client-config claude-code` prints merge-ready settings (also claude-desktop, cursor, codex, kimi-code, pi)
• Organize and layout: merge, split, rotate, delete, reorder, extract, booklet imposition, N-up grids and multi-sheet poster tile slicing
• Optimize and compress: lossy compression with stream deduplication and CMYK safety guards
• Extract and ingest: structured text, detected Markdown tables, Apache Tika / Open-WebUI adapters, and pure-Rust OCR (PaddleOCR models via rten)
• Everything runs locally in pure Rust — the browser app's own pitch is to watch the Network tab and see no uploads

When to use it

Agents that wrangle documents as part of a pipeline — merging reports, compressing scans, extracting tables before analysis — and anyone who wants those operations without uploading PDFs to a service. Privacy-sensitive document work and air-gapped setups fit naturally.

Getting started

# download the paperpilot-mcp binary from Releases, then add it to claude_desktop_config.json:
{
  "mcpServers": {
    "paperpilot": {
      "command": "/path/to/paperpilot-mcp"
    }
  }
}

# from a source build, print a ready-to-merge client config instead:
paperpilot-mcp --client-config claude-code   # also: claude-desktop, cursor, codex, kimi-code, pi

# CLI:
paperpilot merge --inputs doc1.pdf doc2.pdf --output merged.pdf

Things to know

v0.6 (768 stars as of writing, 2026-11-06). Apache-2.0 — the LICENSE file carries the full text even though GitHub's license metadata shows NOASSERTION. Building from source needs Rust 1.75+; Node.js and pnpm are only for the optional desktop app. The compress path is lossy — keep your originals; OCR quality varies with scan quality.