Records screen content and audio 24/7, giving AI a "memory". Users can look back at anything they have seen, said, or used. All data is stored locally, privacy first.
Use Cases
- The user says, "Which webpage did I see that article about… on yesterday?"
- The user says, "Who said what during last week's meeting?"
- The user says, "How much time did I spend on which apps on my computer today?"
- The user says, "Help me find that code snippet I saw earlier"
- The user says, "How much code did I write this week in total? How many emails did I read?"
- Combine with a daily briefing type Skill (if enabled) to automatically generate a daily review based on real screen activity
Prerequisites
- Install Screenpipe:https://screenpi.pe/ （macOS / Windows / Linux）
- After starting Screenpipe, the local API runs at
http://localhost:3030 - Screenpipe has a built-in MCP Server, automatically connected via the MCP protocol
Execution Method
Via MCP tool calls
Screenpipe MCP provides the following core capabilities:
Search screen content (OCR text)
工具: search_screen_content
参数:
query: "搜索关键词"
start_time: "2026-02-25T00:00:00Z" # 可选，时间范围
end_time: "2026-02-26T00:00:00Z"
app_name: "Chrome" # 可选，限定应用
limit: 10
Search audio transcriptions
工具: search_audio_transcripts
参数:
query: "会议讨论内容"
start_time: "2026-02-25T09:00:00Z"
limit: 5
Get app usage statistics
工具: get_app_usage
参数:
start_time: "2026-02-25T00:00:00Z"
end_time: "2026-02-26T00:00:00Z"
Direct API calls (alternative)
If MCP is unavailable, access via HTTP API:
# 搜索屏幕 OCR 内容
curl "http://localhost:3030/search?q=关键词&content_type=ocr&limit=10"
# 搜索音频转录
curl "http://localhost:3030/search?q=关键词&content_type=audio&limit=5"
# 获取最近活动
curl "http://localhost:3030/search?limit=20&start_time=2026-02-25T00:00:00Z"
Typical Workflows
Retrospective Search：
用户：我昨天下午看到一个很好的 Python 库，名字里有 pipe
→ 搜索 OCR 内容，时间限定为昨天下午
→ 返回匹配的屏幕截图和上下文
→ 告诉用户：你在 Chrome 中浏览了 GitHub 上的 xxx 项目（14:32）
Meeting Review：
用户：今天上午的会议讨论了什么？
→ 搜索音频转录，时间限定为今天上午
→ 提取关键讨论点和行动项
→ 结构化输出会议摘要
Time Tracking：
用户：我今天在各个应用上花了多少时间？
→ 获取应用使用统计
→ 生成时间分布报告（可结合图表生成类 Skill 生成图表）
Collaboration with Other Skills
| Combination | Effect |
|---|---|
| screenpipe + daily briefing type Skill (if enabled) | Generate a daily review based on real screen activity |
| screenpipe + meeting-insights-analyzer | Automatically review meeting recordings and screen sharing content |
| screenpipe + habit-tracker | Track habits based on real app usage data |
| screenpipe + pomodoro | Review the actual work content during focus periods |
Output Specifications
- Attach timestamps and source apps to search results
- When screen content is involved, describe the context (which app, which page)
- Annotate speakers in audio transcriptions (if Screenpipe provides speaker labels)
- Present time statistics using tables/charts
- Respect privacy: do not proactively mention sensitive content the user did not ask about