screenpipe

Product Introduction

1. Definition: Screenpipe is a local-first, source-available desktop application (for macOS, Windows, and Linux) that functions as a comprehensive AI agent memory layer. It technically operates as a continuous, multi-modal capture engine and context server, recording on-screen activity, system audio, and application events to create a searchable, timestamped history of a user's digital work.
2. Core Value Proposition: It exists to solve the problem of AI context amnesia by providing AI agents and assistants (like Claude, Cursor, or custom models) with a persistent, detailed memory of a user's computer-based activities. This enables AI to answer questions about past work, automate follow-ups, and generate summaries without requiring users to manually reconstruct their day or provide extensive background context.

Main Features

1. Multi-Modal Local Capture: Screenpipe builds a rich evidence log by combining several capture technologies. It primarily uses the operating system's accessibility APIs for efficient, low-overhead text and event capture from applications. As a fallback and for visual context, it performs GPU-accelerated screen recording with Optical Character Recognition (OCR). It simultaneously captures system audio and microphone input, transcribing it locally or via cloud services. All raw data is stored in a local SQLite database by default.
2. AI-Accessible Context via MCP & Local API: The captured history is not locked in a silo. Screenpipe exposes it through a local REST API (port 3030) and, crucially, a Model Context Protocol (MCP) server. This allows any MCP-compatible AI agent (e.g., within Claude Desktop, Cursor, or Windsurf) to query the user's work history directly, retrieving only the context relevant to a specific prompt or task.
3. Event-Driven Agent Framework ("Pipes"): Beyond passive memory, Screenpipe includes a workflow automation system. Users can create or install "pipes"—YAML-defined agents that trigger based on events (like meeting_ended). These agents can use the local history and connected apps (Linear, Slack, Notion, etc.) to perform actions such as summarizing meetings, updating tickets, or drafting follow-up emails, with full visibility into the trigger, context used, and action taken.

Problems Solved

1. Pain Point: The "context gap" in human-AI interaction. Current AI assistants lack persistent memory of a user's specific work context—past meetings, documents viewed, code written, or decisions made—forcing users to re-explain their situation in every new conversation.
2. Target Audience: Knowledge workers and technical professionals who rely on AI daily, including software engineers, product managers, sales and customer success representatives, founders, and consultants. It is particularly valuable for remote teams and async workers who need to document decisions and context.
3. Use Cases: Recovering specific information ("What did the client ask about pricing last month?"), automating post-meeting workflows (generating notes and action items), drafting periodic updates (weekly summaries from activity), building institutional SOPs from repeated work patterns, and providing deep context to coding assistants about recent bug discussions or feature decisions.

Unique Advantages

1. Differentiation: Unlike cloud-only screen recorders (like Loom or Rewind AI), Screenpipe is local-first and source-available, giving users and enterprises greater control over data privacy and sovereignty. Unlike simple note-taking apps or meeting transcribers, it captures the full cross-application context of work, linking meetings to the documents, code, and messages that surrounded them.
2. Key Innovation: Its integration with the Model Context Protocol (MCP) is a foundational technical advantage. Instead of being a closed AI product, it acts as a universal memory layer for the AI ecosystem, allowing users to leverage their history with any preferred model or agent that supports MCP, future-proofing the investment.

Frequently Asked Questions (FAQ)

1. Does Screenpipe send my screen recordings to the cloud? By default, all capture—screen data, audio, and transcripts—is processed and stored locally on your device. Data only leaves your computer if you explicitly enable a cloud-based feature like cloud AI model usage, team sync, or a third-party integration.
2. How does Screenpipe impact computer performance and battery life? As an always-on capture tool, it uses system resources. Screenpipe employs optimized techniques like accessibility event capture, GPU encoding, and configurable quality/power tiers to minimize impact. Users can expect moderate CPU/GPU and storage usage, which varies based on display resolution, audio transcription settings, and retention policies.
3. Can I use Screenpipe with Claude or Cursor? Yes, directly. Because Screenpipe provides an MCP server, it can be connected as a context source in Claude Desktop or any other MCP-compatible client (like Cursor with MCP enabled). This allows those AI assistants to search your work history within your existing chat interface.
4. Is Screenpipe suitable for enterprise/team deployment? Yes. The enterprise version supports centralized management (MDM, SSO/SAML), device policy enforcement, and audit logs while maintaining a local-first capture architecture on each endpoint. This allows companies to deploy workflow automation without centrally storing all employees' raw screen data.
5. What happens with sensitive information like passwords or private windows? Screenpipe includes configurable privacy controls. You can exclude specific applications (e.g., password managers), window titles, or URLs from capture. It also features automatic redaction filters for sensitive data fields and will pause capture for DRM-protected content.