@khanglvm/tool-hub-mcp

Serverless MCP Aggregator - Reduce AI context token consumption by 38%

Quick Start

# Zero-install usage (recommended)
npx @khanglvm/tool-hub-mcp setup
npx @khanglvm/tool-hub-mcp serve
npx @khanglvm/tool-hub-mcp benchmark

Also works with other JS package runners:

bunx @khanglvm/tool-hub-mcp setup
pnpm dlx @khanglvm/tool-hub-mcp setup
yarn dlx @khanglvm/tool-hub-mcp setup

What It Does

When using multiple MCP servers with AI clients (Claude Code, OpenCode, etc.), each server exposes all its tools to the AI. With 5+ servers, you can easily consume 25,000+ tokens just for tool definitions.

tool-hub-mcp acts as a single MCP endpoint that exposes only 5 meta-tools:

| Tool | Description | |------|-------------| | hub_list | List all registered MCP servers | | hub_discover | Get tools from a specific server | | hub_search | Search for tools across servers | | hub_execute | Execute a tool from a server | | hub_help | Get detailed help for a tool |

Result: 18,613 tokens saved = 38.48% reduction (measured with 6 MCP servers in Claude Code)

Commands

# Import configs from AI CLI tools (Claude Code, OpenCode, etc.)
npx @khanglvm/tool-hub-mcp setup

# Run as MCP server (for AI client integration)
npx @khanglvm/tool-hub-mcp serve

# Add MCP servers manually
npx @khanglvm/tool-hub-mcp add jira --command npx --arg -y --arg @lvmk/jira-mcp

# Run token efficiency benchmark
npx @khanglvm/tool-hub-mcp benchmark

# Run speed/latency benchmark
npx @khanglvm/tool-hub-mcp benchmark speed

MCP Client Configuration

Add to your AI client config:

Claude Code:

claude mcp add tool-hub -- npx @khanglvm/tool-hub-mcp serve

Manual JSON config:

{
  "mcpServers": {
    "tool-hub": {
      "command": "npx",
      "args": ["@khanglvm/tool-hub-mcp", "serve"]
    }
  }
}

Platforms

This package automatically installs the correct binary for your platform:

• macOS (Apple Silicon & Intel)
• Linux (x64 & ARM64)
• Windows (x64)

License

MIT © khanglvm