Blog
Notes on spatial context
Field notes from the Supamaus Lite team — on MCP, screen-aware agents, and working alongside Claude Code and Codex.
10 min read
Accessibility tree vs screenshots: how AI agents should see macOS and iOS apps
Accessibility tree vs screenshots for AI agents driving macOS and iOS apps: what the AX tree contains, real token costs, and why hybrid capture wins.
10 min read
Best screenshot MCP servers for Claude Code and Codex (2026, compared)
Every maintained screenshot MCP server compared — Playwright MCP, Peekaboo, Screenpipe, ios-simulator-mcp and more: platforms, AX trees, pointing, price.
9 min read
Claude Code computer use on macOS: what it does, where it stops, and what to add
Claude Code computer use is live on macOS as a research preview. What shipped, how the screenshot loop works, why small text breaks, and what to add.
10 min read
How to let Claude Code see your Electron app
Electron windows look empty to agents: Chromium skips the accessibility tree until asked. Give Claude Code real visual feedback on your Electron app.
7 min read
Codex can't see your screen? Appshots, MCP servers, and the macOS fix
Codex can't capture your screen by itself. How to get a codex screenshot to the model: Appshots on macOS, the -i flag, paste fixes, and MCP servers.
10 min read
Give Claude Code eyes: let it see your Mac and iOS apps (not just the browser)
Four ways to give Claude Code eyes — pasted screenshots, Playwright, computer use, and spatial context that works on SwiftUI, Electron, and the iOS Simulator.
12 min read
The iOS Simulator feedback loop: build, see, point, verify with Claude Code (2026)
The iOS Simulator MCP options for Claude Code compared — ios-simulator-mcp vs XcodeBuildMCP vs computer use — plus pointing to close the SwiftUI loop.
6 min read
How to paste a screenshot into Claude Code — and how to never need to again
It's Ctrl+V, not Cmd+V — even on a Mac. How to paste a screenshot into Claude Code, fix SSH and VS Code paste failures, and automate capture entirely.
12 min read
Put-That-There (1980) already solved your AI agent's pointing problem
AI agents miss buttons by 40 pixels; HCI solved this decades ago. Fitts's law, Put-That-There, and the bubble cursor mapped GUI grounding before it had a name.
13 min read
What is spatial context? The missing input for AI coding agents
Spatial context is what AI coding agents can't get from your repo: your screen, your pointer, your voice. Here's how it works and how to add it via MCP.