🧩 Auto-Browser: MCP-Native Browser Agent with Human-in-the-Loop

The direction here is clear: giving AI agents access to a real browser while explicitly keeping a human in the loop. Instead of running as a complete black-box automation script, it allows users to intervene or collaborate while the AI navigates web pages. Built as an open-source, MCP-native browser agent, it leverages tools like Playwright and noVNC for local-first, self-hosted deployments. This is a practical piece of infrastructure for developers who need LLMs to handle complex web interactions but want the safety net of manual overrides when the AI gets stuck. The hard part of web automation is rarely the basic clicking and typing; it is handling captchas, dynamic rendering, and unexpected pop-ups. Designing around human intervention is a pragmatic approach given the current limits of autonomous agents.