Annotate

Product Introduction

1. Definition: Annotate is a local-first, macOS screen recording and annotation application designed specifically for generating AI agent prompts. It falls into the technical categories of screen capture software, developer tooling, and AI workflow automation.
2. Core Value Proposition: It exists to eliminate the friction of writing detailed text prompts for AI coding assistants. Its core value proposition is enabling users to communicate intent to AI agents through visual context (screen recordings with drawings) and spoken instructions, which are then handed off directly to compatible AI agents like Cursor, Claude Code, and Codex via Model Context Protocol (MCP).

Main Features

1. Screen Recording with Multi-Monitor Support: Annotate captures high-fidelity recordings of a single application window or across multiple displays simultaneously. This is critical for developers who code on one screen and reference UI on another, ensuring the AI agent receives the complete visual context in one synchronized session.
2. Visual Annotation & Voice Narration: During or after recording, users can draw directly on the screen using pens, rectangles, and text tools to highlight specific UI elements. Concurrently, their voice is recorded and transcribed. The software time-syncs each spoken instruction with the corresponding video frame, creating a rich, multi-modal data stream of gestures and speech.
3. Local-First Session Handoff via MCP: The recorded session (containing video frames and transcript) is processed and stored exclusively on the user's Mac. It is then served to AI coding agents through the Model Context Protocol (MCP), a standard for providing context to LLMs. This allows agents like Cursor AI or Claude Code to "read" the folder containing the session data, interpreting the visual and auditory cues directly, rather than relying on a user's textual description of a screenshot.

Problems Solved

1. Pain Point: The inefficiency and ambiguity of translating complex UI changes, bug reports, or feature requests into written text prompts for AI assistants. Static screenshots lack temporal context, user intent, and procedural detail, leading to guesswork and iterative back-and-forth with the AI.
2. Target Audience: Primary personas include Software Developers and Engineering Managers using AI-powered IDEs; QA Engineers reporting visual bugs; Product Managers communicating design changes; and Solo Founders or Indie Hackers who manage full-stack development and need efficient AI collaboration.
3. Use Cases: Explaining a front-end styling change ("make this button match the sidebar"); demonstrating a multi-step UI bug; walking through a user flow for a new feature specification; providing visual context for refactoring a specific component; creating rich, reproducible prompts for complex coding tasks without typing.

Unique Advantages

1. Differentiation: Unlike generic screen recorders (Loom, QuickTime) which output monolithic video files, Annotate structures data for AI consumption. Unlike typing in an AI chat interface, it provides unambiguous visual proof. Its local-only processing differentiates it from cloud-based recording tools, prioritizing data privacy and security.
2. Key Innovation: The integration of time-synced visual annotation and speech within a local-first MCP server framework. This transforms a simple recording into a structured, queryable context source for AI agents, drastically reducing token waste and increasing prompt fidelity. The focus on AI agent interoperability through MCP, rather than a closed ecosystem, is a significant architectural advantage.

Frequently Asked Questions (FAQ)

1. Is Annotate really free and local-only? Yes, Annotate is completely free for macOS (Apple Silicon) and operates on a local-first principle. All screen recording processing, video frame extraction, and speech-to-text transcription occur on your device. No data is uploaded to external servers, ensuring complete privacy for your work.
2. How does Annotate work with Cursor or Claude Code? Annotate acts as a local MCP (Model Context Protocol) server. After creating a session, you can connect Annotate as an MCP tool within your AI agent (like Cursor). The agent can then read the session's data folder directly, analyzing the synchronized frames and transcripts to understand your request in full context.
3. What are the system requirements for Annotate? Annotate requires a Mac computer with Apple Silicon (arm64 architecture), such as M1, M2, or M3 chips. It does not support older Intel-based Macs or other operating systems like Windows or Linux.
4. Can I use Annotate with any AI, like ChatGPT? Primarily, Annotate is optimized for AI agents that support the Model Context Protocol (MCP) and are designed for code generation, specifically Cursor, Claude Code, and Codex. It is not a general-purpose tool for cloud-based chatbots like ChatGPT's web interface, as they cannot connect to a local MCP server.
5. How is this better than just pasting a screenshot? A static screenshot is a single frame with no indication of process, interaction, or timing. Annotate provides a recorded walkthrough with gestures, clicks, sequence, and your voice explaining the why and how. This gives the AI agent a narrative and precise intent, leading to more accurate and context-aware responses.