PaperPilot gives coding agents local PDF tools with no uploads

A coding agent asked to merge or OCR a PDF usually sends both files to a cloud service. That works for a public document. For a client contract or a medical record, the upload is the reason the agent cannot do the job.

PaperPilot is a pure-Rust PDF toolkit that gives Claude Desktop, Cursor and any MCP client the same operations locally. The post went up on Hacker News and had 79 points by the time of writing. The repository is Apache 2.0, has 155 stars, and was first released on 28 September.

What it does

The author lists the current scope at version 0.6. It can merge, split and rotate pages. It can reorder pages, extract pages, and build a booklet or N-up grid. It compresses. It extracts plain text and runs OCR. It can watermark, validate signatures, remove signatures, run a PII scan, and do a deep sanitize pass. Office data converts from Excel to JSON or CSV. The roadmap after that is PDF to Markdown, HTML and JSON next. Then PDF to Word, Excel and PowerPoint. Then encrypt, decrypt, sign, redact, forms and annotations. Render, flatten, repair, compare and PDF/A are kept for version 1.0.

How it runs

The same Rust engine runs in five places. One is an MCP server for a coding agent. One is a CLI. One is a WebAssembly build that works in the browser with no upload. One is a Tauri desktop app. One is a local REST API on port 7823. The author says PaperPilot uses no Chromium and no C++ runtime. The dependency list has no HTTP client at all, so a session has no network path. The OCR is a pure-Rust port of PaddleOCR through the rten crate.

Where the limits are

Version 0.6 is early software. The README has placeholder lines for the demo gif and the benchmark table. The "benchmarks" section is not yet filled in. Several of the operations many teams need are still on the roadmap. PDF to Word, encryption and signing are not in the Community Edition today. The repository is one month old, so the real-world testing has had little exposure.

For a team deciding whether a coding agent can be trusted with sensitive documents, the trade-off is completeness against privacy. The usual cloud PDF APIs cover far more operations. Each call sends a file. PaperPilot covers the common operations today and keeps the file on the machine. The author released an MCP server next to the CLI. That makes it the first PDF toolkit written for an agent to call directly, with no wrapper in the way.

Source

Primary source: KrushnaVardhanReddy/PaperPilot on GitHub. Discussion: Hacker News thread.