LvcidPsyche/auto-browser

Overview

An open-source MCP-native browser agent that gives AI agents control of a real browser while keeping a human in the loop. It is written in Python and enables interactive web automation with manual oversight.

Best for

Best for
Developers building safe, interactive AI agents for real browser tasks

Use cases

• Automating multi-step web tasks with human approval
• Testing web applications by simulating user interactions
• Extracting data from dynamic websites with visual confirmation

How to use

Install

uvx auto-browser-mcp

Tools exposed

• openai_compatible
• browser.create_session
• browser.observe
• harness.list_runs
• harness.get_status
• harness.get_trace

Tested with

Claude Desktop, Cursor, Continue, ChatGPT, any MCP client that can talk HTTP or stdio

Example client config

{\n  "mcp_servers": [\n    {\n      "name": "auto-browser",\n      "url": "http://127.0.0.1:8000/mcp",\n      "transport": "http"\n    }\n  ]\n}

Notes

An open-source MCP-native browser agent that gives AI agents control of a real browser while keeping a human in the loop. It is written in Python and enables interactive web automation with manual oversight.

531 stars on GitHub. Last updated 2026-05-17. Licensed MIT.

Use cases

• Automating multi-step web tasks with human approval
• Testing web applications by simulating user interactions
• Extracting data from dynamic websites with visual confirmation

Pros

• Open-source and MCP-native for seamless protocol integration
• Human-in-the-loop design reduces risk of unintended actions
• Leverages a real browser for accurate rendering and interaction

Cons

• Requires MCP server setup, adding initial configuration overhead
• Python-only implementation limits language flexibility
• Small community (531 stars) may mean fewer resources and updates

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.