Privacy-first PDF tools for humans and AI agents — entirely local.
Read, search, compare, merge, split, extract, and compress PDFs without uploading documents anywhere.
OpenReader is a local-first PDF utility that works with AI agents.
Every operation — reading, searching, annotating, compressing, comparing, merging, splitting — runs on your machine. No accounts. No subscriptions. No telemetry. No cloud uploads.
Use OpenReader directly, from scripts, or through AI agents.
OpenReader is live in the Microsoft Store.
Recommended: Install OpenReader by Sparsh on Microsoft Store.
For advanced users, MSIX packages remain available on the GitHub Releases page.
| Platform | Package | Notes |
|---|---|---|
| Windows 10/11 | OpenReader.msix |
MSIX package. May be unsigned — requires Developer Mode for sideloading. |
| Windows 10/11 | OpenReader-Setup.exe |
Legacy installer for manual recovery. Requires administrator rights. |
| Windows 10/11 | OpenReader-Windows.zip |
Portable ZIP. |
| macOS | OpenReader-macOS-*.zip |
Experimental. See macOS notes. |
| Linux | — | Unsupported. |
| Platform | Status |
|---|---|
| Windows 10/11 | Supported |
| Microsoft Store | Live in Microsoft Store |
| GitHub MSIX | Advanced users |
| macOS Apple Silicon | Experimental |
| macOS Intel | Experimental |
| Linux | Unsupported |
OpenReader does not install updates itself.
- Microsoft Store installations update automatically through the Store.
- GitHub MSIX installations: Help → Check for Updates opens the releases page. Download and install manually.
- Source builds:
git pulland rebuild.
OpenReader ships with a built-in MCP (Model Context Protocol) server. Any MCP-compatible agent — Claude Code, Claude Desktop, Hermes, or others — can interact with PDFs directly on your machine.
No cloud, no API keys, no document uploads.
| Workflow | What happens |
|---|---|
| Ask questions about a PDF | Agent extracts text from any page and answers. |
| Search entire PDF libraries | Agent indexes a folder and searches across all documents (keyword or semantic). |
| Compare document versions | Agent runs a side-by-side diff and gives you a structured summary. |
| Summarize research collections | Agent reads multiple PDFs and synthesizes findings. |
| Build automated PDF pipelines | Write scripts that merge, split, compress, and extract — all local. |
┌─────────────────────────────────────────────────────┐
│ Claude / Hermes / any MCP-compatible agent │
│ (asks questions, runs searches, compares docs) │
└──────────┬──────────────────────────────────────────┘
│ MCP protocol (stdio or SSE)
▼
┌─────────────────────────────────────────────────────┐
│ OpenReader MCP Server │
│ pdfreader_lib/mcp_server.py │
│ 14 tools: extract, search, compare, merge, split… │
└──────────┬──────────────────────────────────────────┘
│ local file access only
▼
┌─────────────────────────────────────────────────────┐
│ Your PDFs (stored on your machine) │
└─────────────────────────────────────────────────────┘
# Install MCP dependencies
pip install -r requirements-mcp.txtAdd to your MCP-compatible agent's configuration:
{
"mcpServers": {
"openreader": {
"command": "python",
"args": ["-m", "pdfreader_lib.mcp_server"]
}
}
}The server runs over stdio by default. For HTTP/SSE transport:
python -m pdfreader_lib.mcp_server --transport sse --port 8312All operations are local. No data is uploaded anywhere.
| Category | Capabilities |
|---|---|
| Reading | Open PDFs, one-page view, previous/next navigation, page jump, fit-width, zoom in/out |
| Multi-tab | Open several documents in a single window with movable, closeable tabs. Ctrl+T new tab, Ctrl+W close tab, Ctrl+Shift+W close all |
| Session restore | Remembers open PDFs and page positions across restarts. Auto or manual restore (File menu toggle) |
| Search (keyword) | Full-document text search, match count, next/previous result navigation (PageUp/PageDown). Ctrl+F to focus |
| Search (semantic) | TF-IDF cosine similarity search across indexed library. Toggle "Semantic" in search bar |
| Library search | SQLite FTS5 full-text index over entire folders. Cross-document search ranked by BM25. Ctrl+Shift+F shortcut |
| PDF comparison | Side-by-side diff with color-coded changes (red delete, green insert) and diff summary |
| Copying | Drag-select text from the visible page and copy with Ctrl+C or the Copy button |
| OCR fallback | Attempts OCR-assisted selection on scanned/image-based pages when Tesseract OCR data is available |
| Annotations | Highlight, underline, and strikethrough selected text; sticky notes on any page. Saved as native PDF annotations |
| Annotation management | Show/hide annotations toggle (View menu). Delete all annotations on current page or entire document (Tools menu) |
| Save PDF | Explicit File → Save (Ctrl+S) to persist annotation edits immediately |
| PDF tools | Merge PDFs, split every page, extract page ranges like 1-3,5, save compressed copies |
| Dark mode | System-aware dark theme (Catppuccin Mocha) with Auto/Light/Dark toggle via View → Theme |
| Recent files | Quick access to the last 10 opened PDFs via File → Open Recent |
| Update detection | Help → Check for Updates queries GitHub API and opens the releases page. |
| Reader | Sample PDF |
|---|---|
|
 |
| Dark Mode | PDF Tools |
|---|---|
 |
 |
| Sample PDF 2 | |
|---|---|
 |
OpenReader processes PDFs locally. It does not use network services and does not upload PDFs.
The app includes lightweight safety checks before opening and rendering documents:
- Accepts
.pdffiles only. - Checks for a PDF header before parsing.
- Rejects empty files and files over 500 MB.
- Rejects pages outside the supported page-size limit.
- Caps render pixel allocation to reduce PDF-bomb/OOM risk.
- Limits all-pages search result storage.
- Keeps only a small OCR page cache in memory.
- Runs
pip-auditand Bandit in CI.
These checks reduce risk from malformed or oversized PDFs, but PDF parsing still depends on PyMuPDF/MuPDF. Avoid opening PDFs from untrusted sources unless you use OS-level sandboxing, a VM, or another isolation layer.
OpenReader is free software under the GNU AGPLv3.
Copyright © 2026 Sparsh Sam.
python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt
python main.pyBuild the executable:
.\scripts\build_windows.ps1Output:
dist\OpenReader\
├── OpenReader.exe
└── _internal\
├── python311.dll
├── PySide6\
└── ...
macOS packaged builds are experimental. To run from source:
git clone https://github.com/sparshsam/openreader.git
cd openreader
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python main.pySee docs/macos.md for macOS setup and OCR notes.
Text selection works natively on PDFs with embedded text. For scanned/image-only PDFs, the app falls back to OCR via PyMuPDF's Tesseract integration.
Windows: Download Tesseract from UB-Mannheim/tesseract, run the installer, check "Add to PATH", restart the app.
macOS: brew install tesseract
Linux (source builds): sudo apt install tesseract-ocr tesseract-ocr-eng
.
├── .github/ # CI, security checks, Dependabot
├── assets/ # App icon and README screenshots
├── docs/ # Platform notes and known limitations
├── installer/ # Inno Setup installer script (legacy)
├── packaging/ # MSIX packaging
├── scripts/ # Build scripts
├── tests/ # Regression test suite
├── tools/ # Developer utilities and CI test helpers
├── main.py # Main PySide6 application
├── pdfreader_lib/ # Core library (search, comparison, MCP server)
├── requirements.txt # Pinned runtime/build dependencies
├── requirements-mcp.txt # MCP server dependencies (optional)
└── CHANGELOG.md
Contributions are welcome. Please read CONTRIBUTING.md and SECURITY.md before opening issues or pull requests.
| Layer | Choice |
|---|---|
| Language | Python 3.11+ |
| UI Framework | PySide6 (Qt 6) |
| PDF Rendering | PyMuPDF (MuPDF) |
| Search | SQLite FTS5 (keyword), TF-IDF / scikit-learn (semantic) |
| OCR | PyMuPDF / Tesseract integration |
| Packaging | PyInstaller (onedir), MSIX |
| CI/CD | GitHub Actions (Windows + macOS) |
| Security scanning | Bandit, pip-audit |
| Platform | Windows (primary), macOS (experimental) |
Last updated: June 2026