I built mcp launcher because I kept re-pasting the same MCP server configs
into Claude Desktop, then again into Cursor, then again into Windsurf.
@curatedmcp/launcher is one entry you put in any agent's MCP config. It spawns each registered MCP server as a child process over stdio and
aggregates their tools (prefixed with the server's slug). Add a server
once with launcher add <slug> and it works everywhere.
npx @curatedmcp/launcher init
npx @curatedmcp/launcher add github-mcp --env GITHUB_TOKEN=ghp_...
Everything is local — ~/.curatedmcp/stack.json is hand-editable JSON.
No cloud sync, no account, no auth keys. MIT, ~62 KB unpacked, single
dependency (the official MCP SDK).
The catalog of servers comes from a free index at curatedmcp.com that
I've been building. Source: https://github.com/oneprofile-dev/mcp-launcher
What's missing that you'd want? Looking specifically for: failure
isolation feedback, multi-machine sync (or whether anyone wants it),
and whether the slug-prefix on tool names breaks any agent's UI.
Live @ https://www.curatedmcp.com/launcher
Appreciate your thoughts and questions. Thank you, curatedmcp.com
Top comments (0)