---
title: "Connect"
description: "Add a store's customer MCP server to Claude, ChatGPT, Claude Code, Codex, Gemini CLI or any other MCP client."
---

> Documentation Index
> Fetch the complete documentation index at: https://docs.sherpo.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Connect

Connecting takes the store's MCP address and a sign-in. The examples below use `https://shop.example.com/mcp` and the name `shop`: use your store's address instead (see [Where the server is](/customer-mcp#where-the-server-is)), and any name you like. The store's own help page, at `https://{store}/mcp/help`, shows these steps with its address already filled in.

Every client below follows the same flow: you add the address, the client registers itself with the store, and a browser window opens on the store's sign-in page, where you sign in and choose what the client may do (see [Sign-in](/customer-mcp/sign-in)).

### Claude

On claude.ai or in the Claude desktop app:

1. Go to **Customize**, then **Connectors**.
2. Click **+ Add**, then **Add custom connector**.
3. Enter a name, such as `shop`, and the store's MCP address, `https://shop.example.com/mcp`, then click **Continue**.
4. Keep the detected sign-in settings and click **Continue**.
5. Under **Authentication**, choose **Sign in now** (or **Sign in when needed**). Under **OAuth client**, choose **Register automatically**: the store's sign-in takes apps that register themselves, not Claude's published identity.
6. Click **Add**, and approve access on the store's sign-in page that opens.

On a Team or Enterprise plan, an owner adds the connector under **Organization settings**, then **Connectors**, with the same settings, and each member connects it from **Customize**, then **Connectors**.

Claude's own guide: [Get started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp).

### ChatGPT

On ChatGPT on the web:

1. Open **Plugins**, click **+**, and choose **Add custom MCP server**.
2. Enter a name, such as `shop`, and under **Connection** the store's MCP address, `https://shop.example.com/mcp`.
3. Choose OAuth as the authentication, and sign in on the store's sign-in page that opens.
4. Review the warning, choose **I understand and want to continue**, then **Create as a plugin**.
5. Find `shop` in your plugins, or your workspace's, and install it.
6. Start a new conversation, type `@` and pick `shop`.

Whether you can add a custom MCP server depends on your plan and, in a workspace, on what its admins allow.

OpenAI's own guide: [Connecting custom MCP servers in ChatGPT](https://help.openai.com/en/articles/12584461-connecting-custom-mcp-servers-in-chatgpt).

### Claude Code

1. Add the server from your terminal:

   ```sh
   claude mcp add --transport http shop https://shop.example.com/mcp
   ```

2. Start Claude Code and run `/mcp`.

3. Pick `shop` and sign in in the browser window that opens.

Claude Code's own guide: [Connect Claude Code to tools via MCP](https://code.claude.com/docs/en/mcp).

### Codex

1. Open `~/.codex/config.toml` and add the server:

   ```toml
   [mcp_servers.shop]
   url = "https://shop.example.com/mcp"
   ```

2. Sign in from your terminal:

   ```sh
   codex mcp login shop
   ```

Codex's own guide: [Model Context Protocol in Codex](https://developers.openai.com/codex/mcp).

### Gemini CLI

1. Open `~/.gemini/settings.json` and add the server under `mcpServers`:

   ```json
   {
     "mcpServers": {
       "shop": { "httpUrl": "https://shop.example.com/mcp" }
     }
   }
   ```

2. Restart Gemini CLI and sign in:

   ```text
   /mcp auth shop
   ```

Gemini CLI's own guide: [MCP servers with the Gemini CLI](https://github.com/google-gemini/gemini-cli/blob/main/docs/tools/mcp-server.md).

### Other clients

Any MCP client that connects to remote servers over Streamable HTTP and signs in with OAuth works, Perplexity among them. Add `https://shop.example.com/mcp` as a remote server: the client finds the store's sign-in on its own, registers itself and opens the sign-in page. A client that asks for a transport takes **Streamable HTTP** (or just **HTTP**); there is no SSE transport.

Building a client of your own? [Sign-in](/customer-mcp/sign-in#for-developers) has each request it makes.

## After you connect

Ask your assistant something about the store, such as "What's left in my course?" or "Summarize the last lesson I watched". The assistant picks the tools it needs; the tools it sees depend on the access you granted (see [What you can reach](/customer-mcp/availability)).

> **Prompts need a client that supports them**
>
> Tools work in every client above. MCP prompts, which some products offer, are still supported by fewer clients: to be sure you can use them, use Claude (claude.ai, the Claude apps or Claude Code). See [Prompts](/customer-mcp/prompts).

Each connection is to one store. To use the products of two stores, connect each one, under its own name.

## Disconnecting

To cut a client off at once, sign in to the store's site and open **Connected apps** in your account (`/account/connected-apps`): it lists every app connected to your account on that store, with what it may do and when it was last used, and **Disconnect** revokes its access right away. Removing the server or connector in your client also stops it using the store; see [How long a connection lasts](/customer-mcp/sign-in#how-long-a-connection-lasts) for when its access ends then.

Source: https://docs.sherpo.io/customer-mcp/connect/index.mdx
