LvcidPsyche/auto-browser

What it solves

Auto Browser provides a way for AI agents to interact with a real web browser while maintaining human oversight. It solves the problem of "brittle" web automation where agents might get stuck or fail, by allowing a human to step in and take over the session in real-time. It also addresses the challenge of authentication by allowing users to log in once and save those sessions as reusable auth profiles.

How it works

The system uses a Playwright-backed browser control plane that can be driven by LLMs via the Model Context Protocol (MCP), REST APIs, or a dedicated SDK. It consists of a controller API that manages sessions and policies, and a browser node that runs Chromium. A key feature is the integration of noVNC, which allows a human operator to visually watch or take over the live browser session. The system supports various perception presets (like text-only for cost efficiency) and can be driven by any OpenAI-compatible model.

Who it’s for

• Developers building MCP-powered agent workflows that require a real browser.
• Teams needing operator-assisted QA and browser debugging.
• Users managing internal dashboards or admin tools via AI agents.
• Those requiring secure, audited browser automation with PII scrubbing and approval gates.

Highlights

• MCP-native: Built from the ground up as an MCP server for seamless integration with clients like Claude Desktop and Cursor.
• Human-in-the-loop: Live session takeover via noVNC when automation fails.
• Reusable Auth: Save and reopen named authentication profiles to avoid repeated logins.
• Safety and Compliance: Includes approval gates, PII scrubbing, Ed25519-signed Witness receipts for audit trails, and preconfigured compliance templates (e.g., strict, balanced).
• Convergence Harness: A tool for Agent Skill Induction that records and verifies tamper-checked traces to create staged skill candidates.