PaperPilot is a pure-Rust PDF engine that runs locally as a WASM app or MCP server

A PDF toolkit built to stay local

An open-source project called PaperPilot has surfaced on the Hacker News front page with a pitch aimed at the AI-agent era: a PDF and document toolkit written entirely in Rust that runs on the user's own hardware rather than a hosted service. According to the project's GitHub README, one engine backs five delivery surfaces — a CLI, a browser app compiled to WebAssembly, a desktop application, a local REST API and a Model Context Protocol (MCP) server — and documents are processed without ever leaving the machine.

Agent support comes first

The MCP server is the surface the README highlights most. After downloading the paperpilot-mcp binary and registering it in a client's configuration, tools such as Claude Desktop or Cursor can be asked to merge, split, OCR or compress PDFs, with all of the work performed locally. The README presents MCP support as a primary design goal rather than a late addition, which is what separates the project from conventional PDF utilities.

One Rust core, five ways in

Under the hood, a shared Rust workspace — with crates for core logic, PDF handling, NLP, the REST gateway, MCP, WASM and storage — powers every interface so that behavior stays consistent across them:

• an MCP server over stdio for agent clients
• a CLI for batch and scripted workflows
• a browser app at usepaperpilot.com, where files are handled in browser memory and nothing is uploaded
• a desktop app built with Tauri and Svelte, offering a viewer, markup and diffing
• a local REST API served by Axum on port 7823, with Swagger documentation, callable from Python, Node or n8n

The stack carries no bundled Chromium or C++ runtime, and the engine makes no calls to a cloud service, the README states. Building from source requires Rust 1.75 or newer, Node.js and pnpm.

Feature set at v0.6

The current release, v0.6, groups its capabilities into a handful of areas:

• Organization: merge, split, rotate, delete, reorder and extract pages, plus booklet imposition and N-up layouts
• Optimization: compression
• Extraction: plain text and OCR, using a pure-Rust port of PaddleOCR built on the rten runtime
• Security: watermarking, signature validation and removal, PII scanning and a deep-sanitize option
• Office data: Excel files converted to JSON and CSV

The project describes itself as small and dependency-light, built on lopdf, pdf_oxide, rten and related crates, and its benchmark suite can be reproduced with cargo bench.

What is planned next

The roadmap runs through to a 1.0 release. Version 0.7 targets PDF conversion to and from Markdown, HTML and JSON; 0.8 adds export to Word, Excel and PowerPoint; 0.9 covers encryption, decryption, signing, redaction, forms and annotations; and 1.0 is slated to add rendering, flattening, repair, comparison, PDF/A output and Bates numbering.

Licensing and embedding

The repository's Community Edition — which includes the CLI, MCP server, REST gateway, WASM engine and desktop app — is released under Apache 2.0. A separate Enterprise Edition adds SSO, role-based access control, audit logs and data-loss-prevention features. There is also an embeddable widget: a single script tag can drop selected tools such as merge, compress and OCR into any website.

Why it matters

Document handling remains one of the workflows where people routinely upload sensitive files to a web service. PaperPilot's bet is that a fast, pure-Rust engine makes local processing practical everywhere: in the browser via WASM, where users can watch the network tab and confirm nothing is transmitted; on the desktop; and now inside AI agents through MCP, so an assistant can manipulate contracts or scans without a copy reaching a third party. The local REST API extends the same guarantee to automation tools such as n8n. For the wider ecosystem, the pure-Rust OCR path through rten is a notable milestone, since PDF pipelines have historically leaned on C and C++ libraries.