screenpipe records screen and audio locally, converts them into a searchable database, and supports automation through AI agents. It stores data locally by default while offering optional cloud transcription and integration with external assistants.
Project overview
The project combines event-driven capture, OCR, and optional Whisper transcription to create a searchable local history of computer activity with AI agent and MCP server integration.
Project type
MCP · AI Agent
Use cases
Meeting Notes · Audio & Speech
Deployment
Refer to project documentation
License
License pending
Best for
General users and developers who want to capture screen and audio locally and query it without mandatory cloud storage.
Users who want to integrate their screen history with AI assistants via MCP.
Key capabilities
Captures screen and audio by listening for OS events like app switches and typing pauses, pairing screenshots with the accessibility tree and falling back to OCR.
Provides real-time speech-to-text using local Whisper or cloud Deepgram with speaker identification.
Offers full-text and natural language search across accessibility text, OCR, and transcriptions using SQLite FTS5.
Provides a visual DVR-like timeline to scroll through screen history and playback audio.
Allows scheduled AI agents defined as markdown files that query screen data and take actions based on work activity.
Runs as an MCP server allowing AI assistants to query screen history.
Provides a local REST API and JavaScript/TypeScript SDK for searching screen content and audio.
Limitations and risks
The software is source-available under a commercial license. Personal, non-commercial use is permitted, but commercial use requires a license.
Running the application on Linux requires building from source.
Product analytics through PostHog and crash diagnostics via Sentry are enabled by default.
Commercial use of the source requires a paid license.
Getting started
Setup is rated easy because the project provides a desktop installer with auto-updates and allows running via npx without manual build steps. The first success path is downloading the desktop app or running 'npx screenpipe record'.