MCPVault

Actually spawning each server and running the MCP handshake beats scraping READMEs, which is how most directories end up listing servers that no longer start. How do you handle servers that need auth before tools/list returns anything?

Hey DevHunt! Henrik here, maker of MCPVault Launching here fresh off winning Product of the Day on Fazier yesterday and I mean fresh: three launch-thread comments came in, and by the time voting closed we'd shipped 7 commits directly in response to them. That's the energy I want to bring here too, so please let me know what you would like to see, and what I can improve. What MCPVault is: A curated MCP server directory. With Smithery acquired last month, the ecosystem needs an independent, community-first place to discover servers. We ingested 5,000+ from GitHub and built the trust layer that was missing. Every server starts "stale" - just public metadata. Owners claim it, add real docs, and can get verified: we do an actual MCP protocol handshake to confirm the server runs, not just static README analysis. Verified = do-follow backlink + a badge. Grades refresh nightly from GitHub signals. Verified servers get re-tested weekly - fail and the badge lapses automatically. A few things worth calling out: Automated grade disclosure - unclaimed listings (5,100+) clearly state the grade is computed from repo signals alone, no maintainer review. Most directories don't say that. Activity + auth filters - filter by last commit date or auth method (owner-declared, never inferred) ~26K tool sub-pages - so your server gets found even before you claim it Official MCP server - search MCPVault from inside Claude: npx -y mcpvault-mcp Would love to know: what makes you actually trust a directory listing for a tool you're about to give network/file access to?