Screenpipe

Overview

Screenpipe captures screen text and audio into searchable computer history for recall, meeting notes, work summaries and AI context through MCP and a local API. Captured history stays local by default; configured cloud AI, transcription, sync, exports and integrations can transmit context off-device. Source-available under the Screenpipe Commercial License.

From the official site

Capture screen, audio, apps, and meetings locally on macOS, Windows, and Linux. Search your work history or use it with Claude, Codex, and other agents through MCP and API.

The text above is quoted from this tool’s official website — the vendor’s own words.

Official FAQ

What is screenpipe?

screenpipe is AI agent memory: it continuously captures what you've seen, said, and heard on your computer, and gives that context to whatever AI you already use, so it never starts from scratch. It runs on macOS, Windows, and Linux, and exposes your history to any AI through a local API and MCP server.

What will screenpipe actually remember?

screenpipe builds a timestamped, searchable history from on-screen text, app and window context, screen images, system audio, and microphone audio you choose to capture. Use it to find something you saw, reconstruct a day, recover a decision, or draft a follow-up. It cannot recall a period that was not successfully captured.

Do my screen recordings and audio leave my computer?

By default, capture, transcripts, and the search index stay on your device. Data leaves only when a feature you enable needs it, for example cloud AI, cloud sync, an export, a connector, or a team workflow. Each integration has its own destination: local-first does not mean every optional feature runs locally.

Can I exclude sensitive apps, windows, or hours?

Yes. You can pause recording, use recording schedules, include or ignore apps and windows, ignore URLs, and set automatic retention. Privacy filters can redact sensitive text before it is stored, and DRM-protected content pauses capture by default. Review these settings before recording confidential work.

What should I expect from CPU, battery, and storage use?

Any always-on capture uses CPU, battery, and disk. The load depends on your displays, capture quality, audio sources, and local transcription settings. screenpipe uses accessibility events, GPU capture, power tiers, compression, and retention controls to reduce that cost, but it is not zero. If the timeline has gaps, fix capture first. An AI summary cannot recover missing data.

Where can I run it, and can I inspect the code?

screenpipe runs on macOS, Windows, and Linux. Its capture engine is public and source-available under the Screenpipe Commercial License, so you can inspect the code that handles your data, build it yourself, or run the CLI headless.

These questions and answers come from the tool’s own structured data, not written by us.

Key features

• Captures screen text and audio continuously: text is read from apps through accessibility APIs, with OCR as the fallback
• Cross-platform: runs on macOS, Windows and Linux
• Searchable local history: screen, audio, apps and meetings land in a local database you can search by keyword, app or time
• Open to any agent: the local database, REST API and MCP server let AI read that history directly
• Privacy you can narrow: exclude chosen apps, windows and URLs, and redact sensitive fields on the device before storage
• Auditable source: the code is source-available and the data sits in a standard SQLite database, not a proprietary format

How to use it

1. Install the desktop app for macOS, Windows or Linux, or run npx screenpipe record in a terminal — no account required
2. Let it run in the background: screen text, system and mic audio and meeting transcripts keep writing into a local database
3. Search by keyword, app or time, or query the local REST API (port 3030 by default)
4. Point Claude, Codex or your own agent at that history, or set it up so follow-up work runs automatically when a meeting ends

Where it fits

• Meeting notes and action items — When a meeting ends, draft the summary, action items and decisions, then push them to your notes
• Recall and retrieval — Ask what was actually said about pricing last week and get an answer grounded in what you saw and heard
• Context for AI agents — Let an agent read your work history instead of re-briefing it every session
• Turn repeated work into automations — Update the CRM, draft the weekly update or chase follow-ups on a trigger, with the evidence traceable
• A personal work record — Keep projects, people and tasks accumulating in a local knowledge base you own