#!/usr/bin/env node
/**
 * TwelveLabs MCP Server — Main Entry Point
 *
 * ─── What is an MCP server? ──────────────────────────────────────────────────
 *
 * MCP (Model Context Protocol) is a standard that lets an LLM call "tools"
 * that live in an external process. Instead of hardcoding API calls inside the
 * LLM host (like Claude.ai), MCP separates the tool implementations into a
 * standalone server process that Claude talks to over a simple protocol.
 *
 * The flow:
 *   Claude (client) ──stdio──► TwelveLabs MCP Server (this file) ──HTTPS──► ElevenLabs API
 *
 * Claude sends a JSON message like:
 *   { method: "tools/call", params: { name: "twelvelabs_get_agent", arguments: { agent_id: "..." } } }
 *
 * The MCP server receives it, calls ElevenLabs, and streams back:
 *   { content: [{ type: "text", text: "Agent: Tonari-Tutor-v1.2 ..." }] }
 *
 * ─── Transport: stdio ────────────────────────────────────────────────────────
 *
 * We use stdio transport — meaning Claude's host process (e.g. Claude Desktop)
 * spawns this Node.js process as a child and communicates via stdin/stdout.
 * That's why you'll see console.error() instead of console.log() for logging:
 * stdout is reserved for the MCP protocol messages; anything we log must go to
 * stderr to avoid corrupting the protocol stream.
 *
 * ─── Tool registration ───────────────────────────────────────────────────────
 *
 * Each tool is defined in a separate file under src/tools/. We call
 * register*Tools(server) here to attach them all to the McpServer instance.
 * McpServer then handles the protocol machinery (list_tools, call_tool, etc.)
 * automatically — we never write raw protocol handlers.
 */
export {};