How do you choose between competing MCP tools for the same task?

Building an agent pipeline and ran into something I suspect others have hit too. When multiple MCP servers can handle the same task — web scraping, PDF/OCR extraction, invoice parsing, data analysis — how do you actually decide which one to use? Do you benchmark them yourself? Pick based on GitHub stars? Just hardcode whichever you tried first and move on? Genuinely curious whether this is a friction point for others or whether I'm overcomplicating it. Would love to hear how you're actually handling it in p