So IPIPAI just shipped an MCP server, and a lot of you asked how to actually hook it up to your favorite AI tool. Good news: it is simpler than you would think. No local install, no API key juggling, just one endpoint you point at. Here is how I would set it up, and I will walk you through each piece as we go.
First, What Is MCP Anyway?
Short version: MCP (Model Context Protocol) is a standard way for AI apps to talk to external tools and data sources. Think of it as a USB-C port for your LLM. Claude, ChatGPT, Cursor, and a bunch of other clients all speak MCP, so any tool that exposes an MCP endpoint can plug right in. Instead of writing a custom integration for every app, you build one MCP server and every client just works.
Under the hood, MCP runs over JSON-RPC 2.0. That is a lightweight request/response protocol where each message is a small JSON object with a method, an optional params, and an id. When you send a request, you get back a matching response with the same id. Simple, boring, and very easy to debug. If JSON-RPC is new to you, check out this guide on the MCP protocol — it explains the handshake in plain English.
What You Are Connecting To
IPIPAI's MCP server lives at one URL:
POST https://ipipai.com/mcp
That is it. It is a single HTTP endpoint that accepts JSON-RPC 2.0 messages. You POST to it, and it responds. There is no streaming to babysit, no websocket to hold open for the basic tools — a plain HTTP round trip does the job. You can browse every tool the server exposes right now in /mcp-tools.json, and the full reference is on the MCP docs page.
Connecting Claude
For Claude (the desktop app or API setup), you add the remote server in the MCP settings. The short version of the config looks like this:
{
"mcpServers": {
"ipipai": {
"url": "https://ipipai.com/mcp"
}
}
}Save it, restart the app, and Claude should list the IPIPAI tools under the IPIPAI namespace. Now you can ask things like "look up 8.8.8.8" and Claude will call the server on your behalf. That whole flow — tools/list then tools/call — is covered in how AI models call IP tools.
Connecting ChatGPT or a Browser Client
ChatGPT and other browser-based clients have rolled out MCP support too. The idea is the same: you point the client at the remote endpoint. Where Claude wants a settings file, ChatGPT has a built-in "connect a server" flow where you paste the URL and authenticate if the server asks for it. IPIPAI's server is straightforward, so pasting https://ipipai.com/mcp is usually all you need.
Connecting Cursor
Cursor is where this gets genuinely fun for developers, because now your AI pair-programmer can look up IPs live. In Cursor, open Cursor Settings → MCP and add a remote server entry pointing at the same URL. From there, when you are debugging a network issue in the chat panel, Cursor can call IPIPAI's lookup_ip and rate_ip_node tools and drop the real results straight into the conversation. No copy-pasting between tabs.
What Happens Behind the Curtain
No matter which client you use, the actual traffic is the same three moves. The client sends an initialize message, the server answers with its capabilities, then the client calls tools/list to discover what is available and tools/call to run a specific tool. Every one of those is a JSON-RPC 2.0 message posted to the single endpoint. If something ever acts weird, open your client's network log, find the POST to /mcp, and read the JSON — it is hard to get lost.
Quick Sanity Check
Before you trust a client to do it for you, here is the fastest way to confirm the endpoint is reachable: a single tools/list call.
curl -X POST https://ipipai.com/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}You should get back the same tool list that /mcp-tools.json shows you, wrapped in a JSON-RPC response with "id":1. If that works, every MCP client will work. It is the one test I always run first to save myself the headache of debugging a client config that is actually just a network problem.
Where to Go From Here
Once you have it wired up, the server's tools start to matter more than the connection. The full reference for every tool, its parameters, and sample responses lives on the MCP docs page, and the machine-readable list is at /mcp-tools.json. The announcement post, IPIPAI MCP server released, covers the design goals behind it. And if you would rather call the tools from your own code instead of through a chat client, the IP lookup API gives you the same intelligence as a plain HTTP endpoint. Give it a spin and see what your AI assistant does with a live network tool on tap.