Contextberg

Product Introduction

1. Definition: Contextberg is a local AI agent memory application for macOS and Windows. It is a desktop productivity tool that functions as a background system monitor and a Model Context Protocol (MCP) server, designed to augment AI-powered coding assistants.
2. Core Value Proposition: Contextberg exists to solve the "context loss" problem in human-AI collaboration. It automatically captures and structures a user's digital workflow—including screens, browser history, and agent conversations—into a private, searchable memory. This context is then served directly to AI agents like Claude Code and Cursor, eliminating the need for manual context re-entry and enabling seamless task resumption.

Main Features

1. Automated, Multi-Source Context Capture: Contextberg operates silently in the background, continuously capturing data from multiple sources without user intervention. It takes periodic screenshots of all application windows, logs browser navigation history (presumably from supported browsers), and records interactions with integrated AI agents like Claude Code and Cursor. This data forms a comprehensive, timestamped archive of the user's digital activity.
2. Structured, Searchable Memory Synthesis: The raw captured data is processed into three distinct, LLM-optimized memory types. Activity Memory provides a granular, chronological log of actions. Daily Memory aggregates and summarizes progress on a per-day basis. Long-term Memory distills frequently used tools, work patterns, and behavioral tendencies over time. This structure allows AI agents to query for specific, relevant context rather than raw data dumps.
3. Privacy-First Model Routing & Native MCP Integration: All user activity history is stored locally on the user's machine. Contextberg features a built-in MCP server, allowing for zero-configuration integration with MCP-compatible AI agents. Users can flexibly route context queries to different inference endpoints: using an existing Claude desktop app sign-in, external APIs (Gemini, OpenRouter), the optional Contextberg Cloud service, or a fully local model via LM Studio, ensuring data never leaves the local machine if required.

Problems Solved

1. Pain Point: Context Fragmentation and Manual Overhead. Developers and knowledge workers constantly lose time and mental focus re-explaining their current task, open files, browser tabs, and recent steps to their AI assistants after a break or when starting a new session.
2. Target Audience: AI-Augmented Software Developers and Technical Professionals. Primary users are developers who rely on AI coding assistants (Cursor, Claude Code, GitHub Copilot) within their IDE. Secondary users include researchers, data analysts, and product managers who use AI agents for complex, multi-step tasks and need persistent memory across sessions.
3. Use Cases: Seamless Task Resumption: Returning to work after a meeting or break and having the AI assistant immediately aware of the previous code file, error message, and relevant documentation tab. Complex Project Onboarding: Providing a new AI agent with the full context of a week's work, including design decisions noted in browsers and implementation attempts in the IDE. Cross-Session Workflow Continuity: Maintaining context for a long-running feature development or bug investigation across multiple days without manual note-taking.

Unique Advantages

1. Differentiation: Unlike simple clipboard managers or screen recording tools, Contextberg is purpose-built for AI agent integration via MCP. Unlike cloud-based activity trackers, it prioritizes local-first data storage. Compared to manually prompting agents with context, it provides automated, structured, and queryable memory.
2. Key Innovation: Its core innovation is the automated distillation of heterogeneous, real-time user activity (screens, browser, agent chats) into a structured, LLM-queryable memory graph served via a local MCP server. The combination of native macOS/Windows capture, OCR for screenshot search, source exclusions, and flexible model routing creates a unique, privacy-conscious layer between the user and their AI tools.

Frequently Asked Questions (FAQ)

1. Is Contextberg free? Yes, according to the provided schema.org data, Contextberg's offer price is listed as "0" JPY, indicating it is currently a free application for macOS and Windows.
2. How does Contextberg handle my privacy and data? Contextberg is designed with a privacy-first architecture. All screen captures, browser history, and activity logs are stored locally on your computer. You can choose to use a fully local model (e.g., via LM Studio) for processing, ensuring no data ever leaves your device. When using cloud-based models (Claude, Gemini, etc.), only the context necessary for the specific query is sent to the chosen API endpoint.
3. What AI agents are compatible with Contextberg? Contextberg is compatible with any AI agent that supports the Model Context Protocol (MCP). This explicitly includes Claude Code, Cursor, and OpenClaw, as mentioned. Its built-in MCP server means it should work with any other MCP-enabled assistant without complex configuration.
4. What is the difference between Contextberg and a simple screen recorder? While both capture screens, Contextberg adds critical layers: it performs OCR to make screenshot text searchable, structures activity into different memory types (daily, long-term), integrates browser and agent chat history, and—most importantly—serves this processed context directly to AI tools via MCP for actionable intelligence, rather than just producing a passive video log.
5. Can I exclude certain applications or websites from being captured? Yes, the launch notes mention "source exclusions" as a feature. This implies users can configure Contextberg to ignore specific applications (e.g., private messaging apps) or sensitive browser URLs, giving control over what enters the memory archive.