
Auten, which builds an MCP server that gives clients a real screen to operate, published a checklist saying silent macOS failures on Claude Code, Cursor and Codex come from privacy permissions, not the model.
Summaries like this, in your inbox every morning.
The piece draws a line between the two permissions a screen-operating agent needs. Accessibility lets it read other apps' UI trees and send synthesized mouse and keyboard events; without it, Accessibility API calls fail with kAXErrorAPIDisabled (-25211) and events sent via CGEventPost are discarded with no error. Screen Recording lets it capture other apps' pixels, and the write-up notes the less obvious loss: other apps' window titles via CGWindowListCopyWindowInfo's kCGWindowName also become unavailable, so even a search for a window by name fails without a single screenshot being taken.
Where the permission lands is described as the point that consumes the most time. macOS attributes the request to a "responsible" app, which for a command-line tool is usually the GUI app that launched it rather than the binary itself. Auten's own case is instructive: it makes an MCP server that supplies "hands" to operate a real screen for any MCP client, covering Mac, Windows and Linux plus Android phones in beta, with a macOS installer that walks through the permission setup and an auten status command that shows what is connected and what is missing. Silent permission errors, it says, were where the company itself first stumbled.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Vannor launched its agentic marketing model, in which autonomous AI agents research, create, review, publish a…

Microsoft, Alphabet, Meta and Amazon are expected to spend more than $700 billion on AI infrastructure this ye…

A developer released Trading Dojo, an open-source MCP system that hands AI agents like Claude Desktop and Anti…

The developer of HACH, an AI agent resident in Slack / Teams / Chatwork, laid out three things that break duri…

TypeSafe AI released Jev, a System One Model that returns defined decisions instead of free text, with publish…

OpenAI published 722 math research manuscripts made by an unreleased AI model on October 6, 2026; a sign error…
