datboip/HumanAIE
Human AI Eyes (pronounced Human-Eye) — shared browser for human-AI collaboration. Highlight-to-teach builds spatial memory for your AI agent.
Platform-specific configuration:
{
"mcpServers": {
"HumanAIE": {
"command": "npx",
"args": [
"-y",
"HumanAIE"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
Human AI Eyes (pronounced "Human Eye") — a shared browser for human-AI collaboration.
<!-- badges -->
---
AI browser tools are invisible. The agent navigates a headless browser you can't see, gets stuck on a captcha, and sits there burning tokens until it times out. There's no way to watch what it's doing, no way to help, and definitely no way to teach it where to click next time.
HumanAIE fixes that. It runs a headless Chromium instance and streams the viewport to a retro Netscape Navigator-themed UI in your browser. You watch the AI work in real-time. When it hits a captcha or can't find a button, it pings you -- notification chime, browser notification, yellow banner. You click through the captcha, the AI reads the coordinates and keeps going. No copy-pasting screenshots, no back-and-forth describing what's on screen.
The highlight-to-teach system is what makes this different from everything else out there. You highlight an element on the page, and HumanAIE logs those coordinates tied to the URL and a label. Next time the AI visits that page, it already knows where the login button is. You're building a spatial memory for your agent just by clicking.
---
git clone https://github.com/datboip/HumanAIE.git
cd HumanAIE
npm install && npx playwright install chromium
npm startOpen http://localhost:3333/cam/ and you're live.
docker build -t humanaie .
docker run -p 3333:3333 humanaieOne-liner for fresh machines:
curl -fsSL https://raw.githubusercontent.com/datboip/HumanAIE/main/install.sh | bash---
AI Agent <--> REST API <--> HumanAIE Server <--> Headless Browser
Loading reviews...