BrowserOS

by BrowserOS Project

Open-source, privacy-first AI browser that runs agentic automation locally on your machine

See https://browseros.com

Features

  • Chromium-based browser with full Chrome extension compatibility
  • Natural language task automation: run AI agents to click, fill forms, scrape and navigate websites
  • Local-first AI: run local models via Ollama/LMStudio or bring-your-own API keys (OpenAI/Anthropic/etc.)
  • Zero telemetry by default; privacy-preserving local processing
  • Split-view model panel to use multiple LLMs side-by-side (ChatGPT, Claude, Gemini, local models)
  • Semantic search across history and bookmarks
  • MCP (Model-Context-Protocol) support for integration with other AI tools
  • Cross-platform: Windows, macOS, Linux (AppImage/Deb)

Superpowers

BrowserOS turns the browser into an autonomous assistant: instead of manually clicking through pages, you can ask the browser to complete multi-step web tasks in natural language. It’s best for privacy-conscious users, researchers, and power users who want AI automation without sending browsing data to third parties. Running models locally (Ollama/LMStudio) enables full data sovereignty and offline operation.

Pricing

  • Free and open-source.
  • Optional paid cloud LLM usage if you connect provider API keys (costs determined by the chosen LLM provider).

Practical examples

  • “Find the latest quarterly earnings for Company X, extract revenue numbers, and add them to a CSV.”
  • “Open the job listings on example.com and submit my resume to any role with ‘engineer’ in the title.”
  • “Summarize this article and highlight the top 5 takeaways for a presentation.”

MCP (Model-Context-Protocol) role

BrowserOS functions primarily as an MCP server. It exposes an MCP endpoint that MCP-aware clients (LLMs or tooling) can connect to in order to control the browser—running agents, clicking/filling pages, reading page context, and orchestrating multi-step workflows. This lets external clients such as claude-code, gemini-cli, or other MCP clients drive the browser as a controllable agent while the browser itself uses configured LLM backends (OpenAI, Anthropic) or local runtimes (Ollama/LMStudio) to power its automation.

Notes & references