The MCP (Model Context Protocol) gold rush has a retention problem. Developers who eagerly installed a dozen or more MCP servers when setting up Claude Code are quietly deleting most of them within weeks.
The pattern keeps repeating: someone discovers MCP, installs every server that sounds useful, then gradually realizes they only touch three or four of them regularly. After three months of real use, the survivors tend to be the boring ones - filesystem access, git integration, and GitHub's official MCP server. The flashy third-party alternatives to built-in file management? Most people try them and switch back to defaults.
This tracks with how developer tooling adoption has always worked. The tools that stick are the ones that remove friction from tasks you already do every day, not the ones that enable workflows you might theoretically want. An MCP server that lets Claude Code read and write files isn't impressive on a features list, but it's the one you'd notice immediately if it disappeared.
GitHub's MCP server, accessible through claude mcp add, has emerged as one of the most consistently used third-party additions. It connects Claude Code directly to pull requests, issues, and repository data without switching context - the kind of small convenience that compounds over a full workday.
The broader takeaway for the MCP ecosystem: server developers should worry less about feature count and more about whether their tool solves a problem users hit multiple times per day. The MCP servers collecting dust in abandoned configs all share a common trait - they solved problems that sounded important but came up once a week at most.