# screenpipe

Continuous local screen and audio recording with an MCP server and REST API so agents can search what you did.

A Rust desktop app and CLI that captures screen text, accessibility trees, and transcribed audio into a local SQLite database. Agents query it over the Builder's screenpipe-mcp package or the localhost REST API, and scheduled 'pipes' are Markdown prompt files that run an agent against the history.

Alternative to: Rewind

- Capabilities: adaptable, operable
- Vertical: video-streaming
- Shape: product
- Form factor: desktop
- Execution model: hybrid
- Provenance: unknown (recorded, never ranked)
- Repo: https://github.com/screenpipe/screenpipe (Screenpipe Commercial License (source-available; free for personal, non-commercial use)), 21376 stars, last commit 2026-09-03
- Last activity: 2026-09-03
- Page: apps/screenpipe--screenpipe.html
- JSON: apps/screenpipe--screenpipe.json

## Use it when

- I want my agent to answer 'what was that link from this morning's call' from my own machine
- I want a daily standup or meeting summary generated from what I actually did, with the data staying local

## Gotchas

- Source license allows personal, non-commercial use only; commercial use of self-built binaries needs a license
- ~20 GB of storage per month and 5 to 10 percent CPU
- README quotes $25 Standard and $50 Pro while the pricing page shows $21 Basic and $33 Business; the page is newer
- Vertical fit is loose: this is screen recording for agent memory, not production video

## Editions

- Free: Free, open source
  - binary: https://screenpipe.com/how-to-install?download=1
  - installable: npx screenpipe record
  - source: https://github.com/screenpipe/screenpipe
- Basic: $21 per month, closed source
  - hosted: https://screenpipe.com/pricing
- Business: $33 per month, closed source
  - hosted: https://screenpipe.com/pricing
- Enterprise: Free, closed source
  - hosted: https://screenpi.pe/team

## Adaptable: make it yours

Effort: config-only (8 things to edit)

Write a pipe.md under ~/.screenpipe/pipes/ with a prompt, schedule, and YAML frontmatter permissions (allow-apps, deny-windows, time-range, allow-raw-sql) and screenpipe runs a coding agent on it. Exclude apps from audio capture in ~/.screenpipe/audio-exclusions.json. The MCP server reads SCREENPIPE_LOCAL_API_KEY. Changing capture behavior itself means Rust.

- Env vars: SCREENPIPE_LOCAL_API_KEY, SCREENPIPE_AUDIO_EXCLUSIONS_PATH, SCREENPIPE_ENTERPRISE_TOKEN, SCREENPIPE_TEAM_API_URL
- Config files: ~/.screenpipe/audio-exclusions.json, ~/.screenpipe/pipes/<name>/pipe.md
- Prompt files: ~/.screenpipe/pipes/<name>/pipe.md, crates/screenpipe-core/assets/skills/screenpipe-cli/SKILL.md
- Accounts you'll need: screenpipe account (paid plans, cloud AI, sync)

## Operable: what an agent can drive

Operability: native. Interface: mcp at https://github.com/screenpipe/screenpipe/tree/main/packages/screenpipe-mcp (published by the builder, both).

### Reach (partial coverage)

- Search screen text and OCR history: reachable. packages/screenpipe-mcp README: search your screen recordings; REST API on :3030
- Search audio transcriptions and meetings: reachable. packages/screenpipe-mcp README: audio transcriptions; README MCP section: meeting transcriptions
- Get recent context for the current task: reachable. README MCP section: get recent context
- Run raw SQL against the history: reachable. README Developer API: raw SQL access; pipes gate it with allow-raw-sql
- Create and schedule pipes: reachable. README: pipes are pipe.md files in ~/.screenpipe/pipes; CLI skill covers pipes
- Start or verify always-on capture: reachable. README: screenpipe CLI skill covers recorder service, status, and freshness
- Control the computer: reachable. packages/screenpipe-mcp README: 'control your computer with AI'
- Scroll the visual timeline: not reachable. Desktop app view only
- Sync history across devices: not reachable. Paid Business feature done by the app, not an agent tool
- Central team config and permissions: not reachable. Enterprise admin dashboard; team-* tools read, they do not administer

## Code stack

- Languages: Rust, TypeScript
- Frameworks: Tauri, Whisper
- Package manager: cargo
- Database: SQLite
- Hosting target: macOS, Windows, Linux (build from source)

## Model stack

- Providers: local Whisper, Deepgram (optional), Ollama or any local model, screenpipe cloud AI (paid)
- Models: Whisper Large-V3-Turbo
- Keys you need (names only): SCREENPIPE_LOCAL_API_KEY, SCREENPIPE_ENTERPRISE_TOKEN, SCREENPIPE_TEAM_API_URL
- Prompts live in: ~/.screenpipe/pipes/*/pipe.md, crates/screenpipe-core/assets/skills/screenpipe-cli/SKILL.md
- Tools: screenpipe-mcp search and context tools, team-* tools (enterprise)

## Skills

bundled: Screenpipe ships skills as core assets so an agent can query the screen-recording API, chats and CLI; the repo separately carries contributor-only skills under .claude/skills/.
- Paths: crates/screenpipe-core/assets/skills/*/SKILL.md

## Evidence and confidence

- formFactor: README: native .dmg and .exe installers; CLI with npx (confidence 0.95)
- executionModel: README: local Whisper transcription and OCR at runtime; pipes call an agent (confidence 0.85)
- operability: packages/screenpipe-mcp in the same repo, published to npm as screenpipe-mcp; README states the purpose is providing context to agents (confidence 0.8)
- effort: README: pipes are Markdown files with YAML frontmatter; audio-exclusions.json and SCREENPIPE_AUDIO_EXCLUSIONS_PATH documented (confidence 0.8)
- editions: screenpipe.com/pricing: Free $0, Basic $21/mo, Business $33/seat/mo, Enterprise custom (confidence 0.85)
- active: GitHub pushedAt 2026-09-03
- reach:  (confidence 0.75)

## Open questions

- native versus official: the desktop app is the human interface, but the Builder frames agents as the primary consumer; Curator call
- Exact MCP tool names in packages/screenpipe-mcp/src were not read
- Whether this belongs in video-streaming or in a productivity vertical

## Sources

- https://screenpi.pe/ (sourcing:video-streaming:registries)
- https://www.pulsemcp.com/servers/mediar-ai-screenpipe (sourcing:video-streaming:registries)
- https://screenpipe.com/pricing (analysis)