FocusLog: Desktop Activity MCP Server

FocusLog is an MCP server that turns raw desktop usage data into a clean, privacy‑aware activity timeline. It watches which window is active on an X11 Linux desktop, records keyboard and mouse events to compute Actions Per Minute (APM), and compresses this information into a concise log that can be queried by AI assistants. The server is especially useful for developers who want to give an assistant a trustworthy view of their own workflow without exposing sensitive information.

The core value proposition lies in its two‑stage anonymization pipeline. First, a hard‑coded filter removes any user‑specified sensitive keywords from window titles or text snippets. Second, a local LLM running via Ollama performs context‑aware de‑identification of any remaining Personally Identifiable Information (PII). Because the anonymization happens on the client machine, no raw data leaves the user’s environment, addressing privacy concerns that often block AI integration in corporate or regulated settings.

Key capabilities include:

• Real‑time activity logging – captures the focused window title every few seconds and logs keyboard/mouse events for APM calculation.
• APM tracking – provides a metric that correlates with productivity and can be fed into analytics dashboards or AI models.
• Timeline aggregation – groups consecutive similar activities, producing a human‑readable summary that eliminates noise.
• Title sanitization – trims overly long titles and normalizes common patterns, improving readability for downstream tools.
• Configurable operation – log rotation, graceful shutdown, and a dedicated make the server production‑ready.
• Systemd integration – can run as a user service, ensuring persistence across reboots and automatic restarts on failure.

In practice, developers can embed FocusLog into a personal analytics stack: an AI assistant might answer questions like “What did I spend most time on yesterday?” or “Show me a timeline of my last week’s activities.” In corporate environments, the anonymized logs can feed into workflow optimization tools or compliance monitoring without risking accidental data leaks. Because FocusLog communicates over FastMCP, any Claude‑style assistant can request the timeline via a simple prompt, and the server will return structured JSON that the AI can interpret or display.

Unique advantages include its lightweight Linux focus, zero‑configuration LLM anonymization, and the ability to run entirely offline. For developers building privacy‑first AI experiences on Linux desktops, FocusLog offers a ready‑made, battle‑tested source of contextual data that can be leveraged without compromising user confidentiality.