AI app-store engine that records/ indexes your desktop 24/7 and exposes it through a local API so developers can build context-aware AI apps ("pipes").
https://github.com/mediar-ai/screenpipeStop feeding AI models fragments of information. This MCP server connects your AI tools to Screenpipe's 24/7 desktop recording engine, giving your applications complete context about what you've seen, heard, and worked on.
You're building AI applications that need to understand what users are actually doing, not just what they type in a chat. Current approaches involve manual screenshots, copy-pasting text, or asking users to describe their screen. Screenpipe eliminates this friction by automatically capturing and indexing everything - then exposing it through clean MCP tools your AI can actually use.
Complete Desktop History: Every frame, every word spoken, every application used - indexed and searchable through your MCP-enabled AI tools. Your applications get context about what happened 5 minutes ago or 5 days ago.
Real-time Vision Integration: Built-in GPT-4V integration means your AI can literally see what's on screen right now and understand it contextually. No more "take a screenshot and upload it" workflows.
Developer-First Architecture: REST API, WebSocket streams, and TypeScript SDK designed for building production applications. The MCP server exposes these capabilities as standard tools your AI can invoke.
Smart Documentation: AI agents that can automatically document your workflow by understanding what applications you used, what errors you encountered, and how you solved them.
Context-Aware Debugging: When you ask "why did this deployment fail?", your AI can actually look at your terminal history, browser tabs, and monitoring dashboards to give you real answers.
Automated Meeting Notes: AI that watches your screen during calls and generates summaries based on what was actually shared and discussed, not just transcripts.
Personal Knowledge Base: Your AI assistant can reference that config file you edited last week, that error message you saw yesterday, or that design mockup from this morning.
The MCP server runs locally alongside Screenpipe's recorder. Your AI applications connect via MCP and get access to tools like:
screen-capture: Get current or historical framesdesktop-history: Search through past activitiesvision-llm: Analyze screen content with AIaudio-capture: Access microphone recordings and transcripts// Your AI can now do this:
const currentScreen = await tools.screen_capture();
const analysis = await tools.vision_llm({
prompt: "What's the error in this terminal?",
image: currentScreen
});
Built on Screenpipe's proven architecture: Rust for efficient recording (10% CPU usage), TypeScript for the SDK, and native OCR for text extraction. The system stores 15GB per month while maintaining real-time performance.
The MCP server exposes Screenpipe's HTTP API and WebSocket streams as MCP tools, handling authentication and data formatting automatically. Your AI applications get clean, structured access to desktop context without dealing with video processing or OCR implementation.
Ready to give your AI applications the context they need? Install the Screenpipe MCP server and start building applications that understand what users are actually doing, not just what they're saying.