@curatedmcp/launcher

The MCP Hub. One config that bridges every AI agent (Claude, Cursor, Windsurf, Copilot, Gemini) to every MCP server you register.

npx @curatedmcp/launcher init

Plug it in once. Add servers anytime. Use them in any AI agent.

Why

If you use MCP servers across multiple AI clients, you've felt this pain:

• You configure GitHub MCP in Claude Desktop. Then you switch to Cursor and have to do it again.
• You add five servers to Claude. Want them in Windsurf too? Edit a different config file.
• A new AI agent ships? Re-paste every server config from scratch.

Launcher fixes that. It's one MCP entry that fans out to every server you've added, in every AI client.

   Claude   Cursor   Windsurf   Copilot   Gemini
       \      \      |      /      /
        ┌──────────────────────────┐
        │  @curatedmcp/launcher    │   ← one config in each agent
        │  (the MCP hub)           │
        └────┬──────┬──────┬───────┘
             │      │      │
          GitHub  Postgres  Stripe   ← `launcher add`'d once, available everywhere

Install (60 seconds)

1. Add Launcher to your AI client

Drop this entry into your MCP config:

{
  "mcpServers": {
    "curatedmcp": {
      "command": "npx",
      "args": ["-y", "@curatedmcp/launcher"]
    }
  }
}

Config file location:

| Client | Path | | --------------- | --------------------------------------------------------------------- | | Claude Desktop | ~/Library/Application Support/Claude/claude_desktop_config.json (mac) / %APPDATA%\Claude\claude_desktop_config.json (win) | | Cursor | ~/.cursor/mcp.json | | Windsurf | ~/.codeium/windsurf/mcp_config.json | | Claude Code | ~/.claude/mcp.json (or .claude/mcp.json per-project) |

2. Add servers to your stack

npx @curatedmcp/launcher add github
# Prompts for GITHUB_TOKEN

npx @curatedmcp/launcher add postgres --env DATABASE_URL=postgres://...
npx @curatedmcp/launcher list

3. Restart your AI client

Tools appear with a <slug>__ prefix:

• github__create_issue
• postgres__query
• filesystem__read_file

That's it. Add more servers any time — just add and restart.

CLI Reference

launcher                      # Run as MCP server (used by AI clients)
launcher init                 # Print the config snippet for your AI client
launcher add <slug>           # Add a server from the CuratedMCP catalog
  --env KEY=value             # Pre-supply env vars (otherwise prompted)
launcher remove <slug>        # Remove a server from your stack
launcher list                 # Show your stack
launcher --version            # Print version
launcher --help               # Print help

How it works

1. Your AI client launches npx @curatedmcp/launcher over stdio (one MCP entry, like any other).
2. Launcher reads ~/.curatedmcp/stack.json and spawns each registered server as a child process over stdio.
3. On tools/list, Launcher aggregates every child's tools and returns them prefixed with the server's slug.
4. On tools/call, Launcher routes the request to the matching child by name prefix and forwards the response unchanged.

This makes Launcher invisible to the agent — it sees one MCP server with all the tools — while behind the scenes you've got N independent processes, isolated, each with its own credentials.

Config file

~/.curatedmcp/stack.json — plain JSON, hand-editable, version-controllable:

{
  "version": 1,
  "entries": [
    {
      "slug": "github",
      "name": "GitHub",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "ghp_xxxxxxxxxxxx" },
      "addedAt": "2026-05-01T10:14:00.000Z"
    }
  ]
}

Set "disabled": true on an entry to skip it without removing it.

In-agent discovery

Launcher itself exposes 5 discovery tools to your AI client, so you can ask the agent:

"Find me an MCP server for Postgres." "What's the best Stripe MCP?" "Add the Postgres MCP server to my stack."

The agent uses search_servers, get_server_details, and add_to_stack to do all of that without you leaving the chat.

Privacy

• All config is local at ~/.curatedmcp/stack.json. No cloud sync, no account.
• Anonymous telemetry only (event names like "search", "add"). Disable with --no-telemetry or CURATOR_TELEMETRY=false.
• A persistent UUID is stored at ~/.curatedmcp/launcher.json for de-duplication.

Compatibility

• Works with Claude Desktop, Claude Code, Cursor, Windsurf, Copilot, Gemini, OpenAI Agents — anything that supports MCP over stdio.
• Node.js ≥ 18.
• Single dependency: @modelcontextprotocol/sdk.

Links

• 🌐 curatedmcp.com/launcher
• 📚 Marketplace
• 🐙 GitHub
• 💬 Issues

MIT licensed.