Screenpipe – Local-first screen recording with AI memory for agents

TL;DR

TL;DR: Screenpipe runs a continuous screen and audio recorder locally on your machine, transcribes everything, and gives you a searchable history — plus an MCP server so AI agents like Claude Code can query what you’ve seen and heard.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: screenpi.pe ← verified via site fetch
• Source repository: github.com/screenpipe/screenpipe ← custom license, source-available
• License: Screenpipe Commercial License (source-available, not pure OSS) — verified via GitHub
• HN launch thread: news.ycombinator.com/item?id=49024620 — 88 pts, YC S26
• Source last checked: 2026-08-17

What Is Screenpipe?

Screenpipe is a local-first, private screen and audio recording tool that runs 24/7 on your machine. Every frame captured is transcribed and stored in a local SQLite database, making your entire screen history searchable via natural language. The core capture engine and text extraction are open on GitHub; the AI search and cloud sync are optional paid layers.

The defining feature for developers is the built-in MCP server, which exposes your screen and audio memory to any MCP-compatible AI agent. If you’ve ever asked an AI to “check what I worked on yesterday” and gotten nothing useful, Screenpipe solves that by giving agents persistent, queryable memory of your actual work.

The product positioning is “workflow memory for AI agents.” From the site:

screenpipe is the best screen record app for teams. Captures screen and audio, makes everything searchable, and automates workflows. Local-first and private.

Core features from the product page and GitHub README:

• Event-driven screen capture with accessibility-first text extraction
• Real-time audio transcription
• AI-powered natural language search across your screen history
• Local-first privacy: recordings stay in a local SQLite database unless you enable sync
• Cross-platform: macOS (Intel and Apple Silicon), Windows, Linux
• Developer API and MCP server for AI agent integration
• Scheduled tasks (“pipes”): markdown files that trigger automations when events occur

Setup Workflow

Step 1: Install

Download from the project site or GitHub releases:

# macOS (Intel or Apple Silicon)
curl -fsSL https://screenpipe.com/install.sh | bash

# Or download directly from the releases page
# https://github.com/screenpipe/screenpipe/releases

After install, Screenpipe starts a background daemon and opens a web UI at http://localhost:3030.

Step 2: Configure capture settings

The default configuration records everything. Adjust what gets captured from the Settings panel:

# CLI config option to set capture frame rate
screenpipe config set --fps 1  # lower fps = less storage

# Or set capture regions
screenpipe config set --region "1920x1080+0+0"

Audio transcription is enabled by default using the built-in speech recognition engine.

Step 3: Search your screen history

Open the web UI or use the CLI:

screenpipe search "what did I work on at 3pm"
screenpipe search --since "2 hours ago" --query "error message"

Step 4: Connect an AI agent via MCP

Install the Screenpipe MCP server in your AI agent. For Claude Code:

# Using the screenpipe MCP server URL
claude mcp add screenpipe --scope project

Once connected, Claude Code can query your screen history directly:

Ask: "What API did I call in yesterday's debugging session?"
→ Screenpipe retrieves relevant frames and transcripts

The MCP server exposes the following tools:

• search_screen_text(query) — full-text search across captured screen frames
• search_audio_transcripts(query) — search across audio transcriptions
• get_recording_metadata(start_time, end_time) — retrieve metadata for a time range

Step 5: Set up scheduled automations (optional)

Scheduled tasks are defined as markdown files called “pipes”:

# pipe.md
trigger: daily at 6pm
---
Review my screen history for the day and generate a summary.

Pipes can write to your notebook, update tickets in Linear, or post to HubSpot after calls — driven by AI reading your screen and audio memory.

Pricing

From the product’s pricing page (verified 2026-08-17):

| Plan | Price | Notes | |---|---|---| | Basic | $25/month | Core capture, search, and local storage | | Business | $50/seat/month | Team features, cloud sync, admin dashboard | | Enterprise | Custom | MDM deployment, SSO/SAML, custom contracts |

Lifetime licenses purchased before the pricing change remain valid. A free tier with local-only usage is also available.

Security Notes

Screenpipe’s local-first architecture means screen recordings and transcripts stay on your device by default. Data movement occurs only when you explicitly enable sync, cloud AI, exports, connectors, or team workflows. The capture engine source code is public on GitHub so you can inspect what the binary does with your data.

For enterprise deployments, Screenpipe offers a Trust Center and Security Architecture documentation. HIPAA-compliant configurations are available for healthcare use cases.

The product is not open source under an OSI-approved license — it uses a custom commercial license. The source is visible and you can build from source, but downstream distribution rights are restricted.

FAQ

Q: Is Screenpipe open source? A: Screenpipe is source-available under a custom commercial license. The capture engine code is public on GitHub and you can build it from source, but it is not officially an open source project per OSI definitions.

Q: How much storage does continuous recording use? A: Storage depends on fps, resolution, and audio settings. At 1 fps with audio transcription, expect roughly 1–2 GB per day per monitor. The local SQLite database stores transcripts at a fraction of the raw video size.

Q: Does Screenpipe work with Claude Code, Cursor, or other AI editors? A: Yes, via the built-in MCP server. Any editor or agent that supports the MCP protocol can connect. The docs at docs.screenpi.pe cover specific integrations.

Q: Can I use Screenpipe without paying? A: Yes, a free local-only tier exists. Paid plans add cloud sync, team workflows, and enterprise features.

Q: What platforms are supported? A: macOS (Intel and Apple Silicon), Windows, and Linux.

Practical Evaluation Checklist

• [ ] Install and complete first-run setup (daemon + web UI at localhost:3030)
• [ ] Verify screen capture is recording (check the recordings library in the UI)
• [ ] Run a natural language search via the CLI or web UI
• [ ] Connect the MCP server to an AI agent (Claude Code recommended for easy setup)
• [ ] Query the agent about recent screen history — observe whether it retrieves relevant context
• [ ] Test scheduled task: create a pipe that triggers on a specific event
• [ ] Check local SQLite database size after a day of recording to estimate storage needs

Conclusion

Screenpipe is a practical tool for anyone who wants AI agents with real memory of their work. The local-first design is the key differentiator — your screen history is yours, not uploaded to a third party by default. For solo developers and small teams who want AI coding assistants that actually know what they’ve been working on, it fills a real gap.

The MCP integration is the most immediately useful part: instead of copy-pasting context into every new session, you can ask your agent to check what you did yesterday or last week. The scheduled automation feature goes further for power users who want routine tasks handled automatically.

The main trade-off is that Screenpipe is not open source (source-available only), and the lifetime of local recordings depends on your storage budget. If those constraints are acceptable, it’s worth trying the free tier.