brightdata/brightdata-mcp ? reverse-engineered prompt

Build me a Node.js MCP server that gives AI agents easy access to public web data.

I want it to let a client search the web, scrape pages into clean markdown or HTML, pull structured JSON from popular sites like Amazon, LinkedIn, YouTube, Reddit, X, and similar, and also control a remote browser for simple actions like navigate, click, type, take screenshots, and read page content. It should also support fetching package info for npm and PyPI, and collecting answers from major AI chat services as structured data.

Make it work as both a local server I can run with npx and a hosted endpoint style setup, with clear config for API token based auth. It should feel reliable on hard to access sites, so handle bot blocking, CAPTCHA, proxy rotation, and retries behind the scenes. Please include a clean README, a few example configs for common MCP clients, and basic tests so I can see it all working. If you need to check current docs for MCP or browser automation, look them up online.