What Happened
mcpup, an open-source CLI tool for managing MCP (Model Context Protocol) servers across multiple AI clients, appeared on Hacker News on March 6. The tool solves a specific pain point: if you use more than one AI coding assistant, you're maintaining separate MCP configurations for each one.
mcpup stores server definitions in a single canonical file at ~/.mcpup/config.json and syncs them to the native config formats of 13 supported clients: Claude Code, Cursor, Claude Desktop, Codex, OpenCode, Windsurf, Zed, Continue, VS Code, Cline, Roo Code, Amazon Q, and Gemini.
The tool ships with 97 built-in server templates across categories like developer tools, productivity, and databases. It supports both local stdio servers and remote HTTP/SSE connections.
Key features include named profiles (group servers for different workflows), backup and rollback before config writes, drift detection when configs change outside mcpup, and a doctor command for diagnosing issues. It preserves any manually-added entries in client configs that it doesn't manage.
Installation is available through Homebrew, Go install, or binary downloads from GitHub Releases.
Why It Matters
MCP adoption has hit a tipping point. Most AI coding tools now support it, and power users are connecting multiple servers - GitHub, Slack, databases, file systems, custom internal tools. The problem is each client stores its MCP config differently, and keeping them in sync manually is tedious and error-prone.
If you use Claude Code at the terminal and Cursor in your IDE (a common setup), you're editing two separate config files every time you add or update an MCP server. Add Windsurf or Continue to the mix and it gets worse. mcpup reduces that to one change, one sync.
The 97 built-in templates also lower the bar for getting started. Instead of hunting down the right server configuration for common services, you can add one with a single command and your credentials.
Our Take
This is a utility that shouldn't need to exist but absolutely does. MCP's rapid adoption across clients has created a fragmentation problem that the protocol spec itself doesn't address. Someone had to build the config sync layer, and mcpup looks like a solid first attempt.
The profile feature is the standout. Being able to define a "work" profile with your company's internal MCP servers and a "personal" profile with different ones, then switch between them, solves a real workflow problem for people straddling multiple contexts.
The 13-client support list is impressive and covers practically every AI coding tool that matters right now. The fact that it preserves unmanaged config entries is a thoughtful touch - it means you can adopt mcpup incrementally without breaking anything.
One concern: config management tools are only useful if they stay current. AI clients update their config formats regularly, and mcpup will need to keep pace. Worth watching whether the maintainer can sustain that across 13 targets.
If you're running MCP servers in more than two clients, this saves real time. Install it, run mcpup doctor, and consolidate your configs.