Local-first knowledge capture with AI context for macOS
Capd, by Jamie Davenport, is a privacy-focused native macOS application that captures and indexes local notes and research as a private second brain. It stores data locally in a SQLite library and exposes local context to AI assistants via the Model Context Protocol for cited answers. The app supports full-text search and image text extraction, and it is aimed at knowledge workers and AI power users who need confidential, locally stored context for their assistant workflows.
How the MCP bridge changes AI-assisted research
The app hosts a built-in Model Context Protocol server that exposes read-only search and cited-answer tools, allowing compatible assistants to query the local library and return sourced snippets. That architecture lets an external assistant reference on-device material without uploading the library to a third-party cloud, and the documentation names compatible assistants explicitly, so the tool's primary job is supplying private context to external models.
How reliable and fast is on-device retrieval?
Capd is written with native macOS frameworks and emphasizes high-performance capture and retrieval on a single machine. The implementation uses Swift and SwiftUI for a desktop experience and supports both Apple Silicon and Intel processors, which the developer cites as design targets. Those choices are intended to reduce latencies for local queries and large-library searches compared with non-native ports.
How does it handle privacy and external access?
The product follows a zero-telemetry, zero-account privacy model and keeps all user data on the device by design. A secure local MCP server exposes read-only query endpoints so assistants can access context without cloud storage. That pattern preserves local control while enabling AI-assisted answers, but it explicitly restricts remote write access because the exposed interface is read-only.
Is it practical to slot into macOS workflows?
Integration options target scripted and hotkey-driven workflows: default hotkeys exist for capture and search, a CLI provides stable JSON output for automation, and a Raycast extension offers launcher-based commands. The app is distributed as a notarized .dmg and requires a recent macOS release, so installation and scripting suit single-machine power users and those who automate captures from other macOS tools.
A focused choice for privacy-minded, single-machine researchers
Capd is a practical option for users who prioritize on-device data control and want their private library to inform external assistants. The developer emphasizes open-source, privacy-first design, and the app has found favor among privacy-conscious communities, making it a sensible pick for individual researchers and AI power users who keep their primary workflow on one Mac.
