Screeny is a privacy-first, macOS-only MCP server that enables AI agents to obtain visual context by capturing screenshots of pre-approved application windows. It enforces Screen Capture permission and a Window Approval setup so only user-approved windows can be captured, in the background, without changing window focus. All processing stays on the local machine and there are no external connections, with screenshots deleted after use. The server exposes two MCP tools: listWindows, which lists only user-approved windows, and takeScreenshot, which captures a screenshot of a window by its ID and returns a full-pixel JPEG image with a base64 payload. By default the base64 cap is Medium (250KB) and can be configured and clamped to 100–900KB. Users can choose presets Tiny, Small, Medium, Large, or XL during setup, stored in ~/.screeny/config.json as max_b64_kb, with an optional SCREENY_MAX_B64_KB environment variable. Approvals are saved to ~/.screeny/approved_windows.json, and you can re-run setup to update the list. A screeny://info endpoint provides server information, and the README shows integration examples for MCP clients like Claude Desktop and Cursor.