What problem does it solve? Connecting AI agents to many MCP servers floods the context window with dozens of tool schemas and requires managing separate endpoints for each server. This Skill provides a single gateway that federates multiple MCP servers behind one endpoint and exposes lightweight meta-tools for dynamic tool discovery and routing. ## Core Features & Use Cases - Meta-Tool Routing: Discover, inspect, and call tools across all connected servers using four meta-tools (gateway_search_tools, gateway_get_tool_schema, gateway_call_tool, gateway_add_server) with under 200 prompt tokens of overhead. - Dynamic Server Registration: Add new MCP servers at runtime by SSE URL, npm package, or GitHub repository link without restarting the gateway. - Three Operating Modes: Choose meta mode for context-optimized agents, hybrid mode for production with ChatGPT connectors, or direct mode for full namespaced tool exposure. - Use Case: An agent needs to screen stocks and then generate a video summary. It searches the gateway for a screener tool, fetches its schema, executes it against the Tickertape server, then routes a video generation call to the Google Flow server, all through one SSE endpoint. ## Quick Start Connect your agent to the gateway SSE endpoint and ask it to search for the tool you need, fetch its schema, and execute it through the gateway.
