jamiedavenport/capd

CAPD is a polished local-first macOS app for saving useful content and finding it later. It can capture web pages, selected text, notes, links, images, and dropped files. Users can search titles, article text, notes, tags, and text found inside images. The project also includes a command-line tool, Raycast extension, share extension, automatic tagging, and read-only MCP access for AI assistants.

Privacy is one of the repository’s clearest strengths. CAPD does not require an account, subscription, cloud service, analytics system, or telemetry connection. Captured content is stored in a local SQLite database and asset folders. OCR, automatic tagging, and library answers use Apple’s on-device tools. The privacy guide clearly explains page fetching, favicon requests, update checks, contextual reminders, link attribution, and MCP data sharing. It also tells users which network features they can disable.

The code is organized into focused Swift packages for the main app, user interface, capture engine, background agent, command-line tools, and share extension. This separation should make the project easier to maintain as it grows. The repository also includes a large and thoughtful test suite. It covers capture behavior, full-text search, database migrations, duplicate handling, URL rules, browser extraction, OCR pipelines, permissions, command output, MCP tools, SQLite write contention, and search performance.

The development process looks healthy. Current CI, documentation, and release workflows are passing. CI runs formatting checks, the full Swift test suite, Raycast checks, and a dedicated search performance test. Release builds are signed, notarized, and distributed through GitHub and Homebrew. The MIT license, contribution guide, installation documentation, screenshots, and short command examples make the project easy to explore.

The biggest area for improvement is security documentation. CAPD runs outside the macOS App Sandbox and can request Accessibility and Automation permissions. Those choices are explained, but they deserve a dedicated threat model and security reporting policy. The documentation should explain what each permission allows, which processes receive it, how the background agent is installed, and how users can remove every helper and permission during uninstall.

Local storage also deserves more guidance. “Stored locally” does not always mean “encrypted.” The docs should explain whether the database and captured files rely on FileVault, whether sensitive captures can be excluded, and how backups or exports should be protected. An optional lock, private collection, or encrypted storage mode could help users who save confidential work material.

MCP access is read-only, which is a good default, but returned text can still leave the Mac through the connected AI client. A preview of what will be shared, clear result-size limits, and per-client access controls would strengthen this feature. Tests for malicious page text, prompt injection inside saved captures, unsafe file paths, oversized images, and hostile MCP requests would add another useful safety layer.

Finally, the project requires macOS 26, which keeps the implementation modern but limits the number of people who can try it. The README should briefly explain why that version is required and whether support for older systems is planned. Since CAPD is still at version 0.0.7, labeling it as an early release and publishing a short stability policy would also set clear expectations.

Overall, CAPD is a focused, attractive, and carefully tested project with a strong privacy story. Its local-first design, fast search, useful capture options, and open documentation make it stand out. A formal security policy, clearer permission guidance, and stronger protection for sensitive local data would make an already promising app even more trustworthy.

Capd is a native macOS menu bar app built for local-first web clipping, notes, and image indexing. Unlike bookmarking services that depend on cloud synchronization and user accounts, Capd keeps your library strictly on your Mac with zero telemetry, external servers, or tracking.

It pairs SwiftUI and AppKit with SQLite FTS5 for instant full-text search, while Apple's Vision framework handles on-device OCR for images. A notable addition is the built-in CLI and read-only MCP server, which makes your captured content accessible to local LLM tools and command line scripts.

Full-text search queries return results instantly, image text recognition runs locally, and the background resource footprint stays light. For Mac users looking for a fast, private, and subscription-free web archive and snippet tool, Capd is a solid, well-crafted app.