Getting started

Welcome to BrowserControl — the vision-first browser automation MCP server. This section walks you from a fresh install to a working agent in under five minutes.

What you'll do

1. Install BrowserControl — pip / uv / pipx in one command.
2. Connect it to your AI — Claude Desktop, Cursor, Cline, Continue, Zed, or programmatic Python.
3. Run your first session — navigate, click, type, and screenshot.
4. Add the agent skill (optional) — a playbook that teaches your agent to drive the browser well.

What you'll need

• Python 3.11 or newer on Linux, macOS, or Windows.
• An MCP-compatible AI client — Claude Desktop, Cursor, Cline, Continue, Gemini CLI, or any client that speaks MCP stdio.
• No API key, no cloud account, no telemetry. BrowserControl runs 100% on your machine.

Already have a Python project?

If you're using uv, BrowserControl drops in as a normal dependency. No browser binary to manage — Chromium auto-installs on first run.

Quick taste

Once connected, ask your agent:

"Open Hacker News and tell me the top story."

The agent will call the MCP tools:

→ navigate_to("https://news.ycombinator.com")
→ get_page_content()
→ "The top story is: ..."

That's the whole loop. No selectors, no XPath, no DOM debugging — just point at numbers.

Where to next

• Install BrowserControl

  ---

  Pick your installer and get Chromium auto-provisioned.

• Connect your AI

  ---

  Drop-in configs for every major MCP client.

• Run your first session

  ---

  A guided walkthrough of the most common tools.

• Add the agent skill

  ---

  A drop-in playbook that makes agents markedly more reliable.