Client Setup & Integration Overview

The Model Context Gateway (MCG) enables AI coding assistants, IDEs, autonomous agents, and custom software to securely connect to downstream MCP servers through standardized, authenticated endpoints.

---

🔌 Connection Endpoints & Routing Modes

MCG provides three primary routing paths for connecting MCP clients:

Route Path	Mode	Description	Recommended For
/sse	Unified Meta-Mode	Exposes only search_tools and execute_tool. Searches the catalog dynamically on demand to minimize context saturation.	All AI Assistants & IDEs (Cursor, Claude Desktop, Cline, Antigravity)
/{targetServerId}	Target Server Proxy	Bypasses Meta-Mode, proxying 100% of the tools, resources, and prompts from a specific server (e.g. /docker, /homeassistant).	Dedicated agents, single-purpose workflows, or clients requiring direct tool schemas.
/admin/sse	Admin Management Mode	Exposes the complete suite of gateway administration tools (manage_servers, manage_appkeys, manage_policies).	Autonomous DevOps agents and admin scripts.

---

🎛️ Dynamic Client Configuration Generator

The Client Setup Guide card (available on both the Overview and App Keys & Security tabs) features an interactive configuration generator:

1. Select Target Route: Choose Unified Meta-Mode (/sse) for aggregated discovery, or choose a specific downstream server (e.g. /docker).
2. Select Client Tool: Choose your AI environment (Cursor, Claude Desktop, Cline / VS Code, Antigravity CLI, or SDKs).
3. Include Key: Check the box to auto-fill your active AppKey into the generated JSON or bash snippet.
4. Copy & Paste: Click Copy and paste directly into your client configuration file.

---

🔐 Client Authentication

All client requests must authenticate using an authorized AppKey:

• HTTP Header (Standard):

  Authorization: Bearer mcp-usr-Xk9L2mPq-7vN3wZ8aB1cE4fG9

• Query Parameter Fallback: If your client or proxy environment cannot send custom HTTP headers during the initial SSE handshake, MCG accepts:

  http://localhost:8080/sse?apiKey=mcp-usr-Xk9L2mPq-7vN3wZ8aB1cE4fG9

---

📚 Dedicated Client Configuration Guides

Follow the step-by-step guide for your preferred AI client:

• Cursor IDE Setup — Configure .cursor/mcp.json for Meta-Mode and direct proxy routing.
• Claude Desktop Setup — Configure claude_desktop_config.json using the @modelcontextprotocol/client-sse bridge.
• Cline, Roo Code & VS Code Setup — Configure cline_mcp_settings.json and VS Code extensions.
• Antigravity CLI & Autonomous Agents — Command-line flags and environment variables for autonomous agent execution.

---

💻 Programmatic SDK Integration

For custom software, scripts, or autonomous agent frameworks, use the official MCP SDKs:

1. TypeScript SDK (@modelcontextprotocol/sdk)

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { SSEClientTransport } from "@modelcontextprotocol/sdk/client/sse.js";

const transport = new SSEClientTransport(
  new URL("http://localhost:8080/sse"),
  {
    requestInit: {
      headers: {
        "Authorization": "Bearer mcp-usr-Xk9L2mPq-7vN3wZ8aB1cE4fG9"
      }
    }
  }
);

const client = new Client({ name: "custom-ts-agent", version: "1.0.0" }, { capabilities: {} });
await client.connect(transport);

// In Meta-Mode, search for relevant tools dynamically
const searchResult = await client.callTool({
  name: "search_tools",
  arguments: { query: "restart container" }
});
console.log("Discovered tools:", searchResult);

2. Python SDK (mcp)

import asyncio
from mcp import ClientSession
from mcp.client.sse import sse_client

async def main():
    headers = {"Authorization": "Bearer mcp-usr-Xk9L2mPq-7vN3wZ8aB1cE4fG9"}
    async with sse_client("http://localhost:8080/sse", headers=headers) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()
            tools = await session.list_tools()
            print("Connected! Available bootstrap tools:", [t.name for t in tools.tools])

asyncio.run(main())