perplexityai/modelcontextprotocol — reverse-engineered prompt

Build me a TypeScript MCP server that lets AI tools use the Perplexity API for live web answers. I want it to work in tools like Cursor, Claude Code, Claude Desktop, Codex, VS Code, Kiro, and Windsurf, with simple setup using a Perplexity API key.

It should give the assistant four abilities, search the web and return ranked results, ask quick everyday questions with live sources, do deeper research reports, and handle harder reasoning tasks. Please include the option to hide thinking text for the deeper research and reasoning responses.

Make it easy to run locally with an MCP config, and also support an HTTP server mode for cloud or shared use. Include environment settings for the API key, timeout, base URL, log level, port, allowed origins, and proxy support for people on corporate networks. Add Docker support, clear README instructions, basic troubleshooting, and tests where they make sense. Look up current MCP and Perplexity docs online if you need to.