# Add a remote MCP server with a URL and a sign-in

> What is a remote MCP server? One you add by URL with nothing to install. See how it differs from a local one and add Tracy's to Claude or ChatGPT.

- URL: https://tracyinterview.ai/blog/what-is-a-remote-mcp-server
- Published: 2026-11-19
- Author: Louis

> **In short** What is a remote MCP server? A program that runs at a web address, so you add it to an AI app by URL and sign in, with nothing to install. **Tracy AI Interviewer**'s is at tracyinterview.ai/mcp: add it to Claude or ChatGPT and the chat can draft, invite and read results while Tracy runs the interviews. You approve each create and invite, and your team decides what the answers mean.

## What is a remote MCP server, and what is it good for?

A remote MCP server is a program that gives AI apps tools and data, and runs on the internet at a web address instead of on your own computer. MCP is the Model Context Protocol, the open standard AI apps use to connect to outside systems. In the protocol's own terms, "MCP server" refers to the program that serves context, regardless of where it runs. A remote one is hosted by someone else and reached by URL.

That design does three things well: there is nothing to install, one hosted copy can serve many apps, and a sign-in can tie each request to the person's own account. If the basic idea is new, [what an MCP server is](/blog/what-is-an-mcp-server) covers it first.

A URL gives the chat its tools. It does not give the chat a voice on a call with the person you want to hear from, and that part is what Tracy takes over once the server is connected.

![Glossary card: remote MCP server, an MCP server hosted at a web address. Example: Tracy's is tracyinterview.ai/mcp.](/blog/images/what-is-a-remote-mcp-server-1.png)

*The definition in one card, with Tracy's server address as the example. Illustration with sample content. Source: Model Context Protocol architecture overview.*

A remote MCP server is hosted at a web address and added to an AI app by URL.

## Remote vs local MCP server

The protocol defines two ways for a client and a server to talk, and the choice follows from where the server runs.

How a local and a remote MCP server differ.

| Question | Local server | Remote server |
| --- | --- | --- |
| Where it runs | On your computer | At a web address |
| Transport | Stdio, standard input and output | Streamable HTTP |
| Who it serves | Typically one app | Typically many apps |
| How you add it | Point the app at a program to start | Paste its URL |
| Sign-in | Credentials come from your environment | Usually OAuth in a browser |

The MCP documentation says remote servers can use bearer tokens, API keys or custom headers and recommends OAuth to get the token. The specification says a server using stdio should read credentials from the environment instead.

## How an AI app connects to a remote MCP server

You give the AI app the server's address and it does the rest: it connects, asks the server what tools it offers and, if the server requires it, sends you to sign in. Each app adds a few conditions of its own.

- **Claude** connects from Anthropic's cloud, so the server must be reachable over the public internet, and Claude cannot connect to a URL that names another port. Claude Code recommends HTTP for remote servers and calls SSE deprecated.
- **ChatGPT** needs a public HTTPS endpoint, or a Secure MCP Tunnel. Public endpoints support streamable HTTP, typically at /mcp, and SSE is also supported.

A server on your laptop is therefore not a remote server. Claude on the web and ChatGPT cannot reach it unless you expose it to the internet. Claude Desktop and Claude Code run local servers, which is why the local kind still exists.

![Four steps: paste the server URL, sign in, approve the preview, Tracy interviews.](/blog/images/what-is-a-remote-mcp-server-3.png)

*From URL to interview: the AI app takes the server URL, you sign in and approve the preview, and Tracy interviews. Illustration with sample content. Source: Claude and OpenAI documentation, 14 Nov 2026.*

Adding a remote server takes an address, a sign-in and an approval, and then Tracy runs the interview.

## A remote MCP server URL, with one example

Tracy AI Interviewer's remote MCP server lives at `https://tracyinterview.ai/mcp`. Paste that address into Claude's custom connector dialog or ChatGPT's custom MCP server form, sign in to Tracy and approve access. The chat can then draft an interview, invite people and read results, and each create or invite waits for your approval of a preview. [What an MCP server is](/blog/what-is-an-mcp-server) lists the tools in full. The setup is covered step by step for [Claude](/blog/tracy-claude-integration) and [ChatGPT](/blog/tracy-chatgpt-integration).

## What Tracy adds to a remote MCP server

### 01. Reach every person

The server address stays with the owner. The person being interviewed starts on click, at any hour, with no scheduling and no recruiter on the call, and Tracy runs the interview on video.

### 02. Ask the next question

Tracy follows the answer in the person's own field and waits for it. A vague reply gets a sharper question before the interview moves on.

### 03. Keep the evidence

Every result holds a recording, a timestamped transcript and a written assessment per question with its reasoning. Tracy reports what people said, and your team decides what it means.

## What a remote server lets a team do, and on whose account

Adding a remote server changes who can act on which account, and the answer depends on the app.

In Claude, a custom connector on a Team plan is added by an owner first, and members then connect it themselves. On Enterprise, a custom role with Manage access to Libraries can also add one. Each member signs in on their own, so the server sees a request from that person's account, not a shared one. In ChatGPT, workspace permissions and security restrictions, including Lockdown, apply to adding and using custom MCP servers, so an admin's settings can stop a server before anyone connects it.

Tracy's server shows what this looks like in practice. It asks for one permission, named tracy, and acts only on the account of the person who signed in. Count it up: one address, one permission and one sign-in per person, so each request comes from the account of whoever signed in. A person can cut it off under Connected AI apps in their Tracy account, and access stops at once. They can connect again later.

Two things sit outside the connection. What a remote server stores about your requests is up to whoever runs it, so read its privacy terms the way you would for any hosted service. And the AI app is still the one asking you before a tool runs, so a team can keep write actions on approval even where it allows the connector.

## How to tell whether a server is remote

Look at what you are asked to give the AI app.

- If it asks for a web address, usually starting with https:// and often ending in /mcp, the server is remote. You paste the address, sign in and you are done.
- If it asks for a command to run, or a path to a program on your computer, the server is local. The AI app starts it, and it stops when the app does.
- If the address is something like localhost, it points back at your own computer. Claude on the web and ChatGPT cannot reach it, because they connect from the vendor's cloud and need a public address.

When a remote server fails to connect, the usual causes are an address typed with a typo, a sign-in that was not completed, and a server that is not reachable from the public internet. Check them in that order before you suspect the server itself.

## Are remote MCP servers safe, and which should you choose?

A remote server runs under someone else's control, so the questions are who runs it and what it can do for you. Anthropic and OpenAI both say to connect only to servers you trust. Read the permission screen when you sign in. In Claude and ChatGPT you can see which tools a server offers and, for write actions, you are asked before they run.

Choose a local server for tools that work on files on your own machine. Choose a remote server for a hosted service, such as Tracy, that more than one person or app needs to reach.

Which path fits you:

- Many applicants to screen: see [hiring at volume](/use-cases/volume-hiring).
- Customer interviews: see [customer research](/use-cases/customer-research).

## Common questions

### What is a remote MCP server?

An MCP server hosted at a web address instead of on your computer. An AI app adds it by URL and usually signs you in.

### What is the difference between a remote and a local MCP server?

A local server runs on your computer over stdio and typically serves one app. A remote server runs at a web address over Streamable HTTP and typically serves many.

### What is a remote MCP server URL?

The HTTPS address where the server accepts MCP requests, often ending in /mcp. Tracy's is https://tracyinterview.ai/mcp.

### Do remote MCP servers need a sign-in?

Not always. The protocol makes authorization optional, but MCP recommends OAuth for remote servers, and a server that acts on your account has to know whose account it is.

### Can Claude on the web or ChatGPT use a server on my own computer?

Not directly. Claude on the web connects from Anthropic's cloud and ChatGPT needs a public HTTPS endpoint or a Secure MCP Tunnel. Claude Desktop and Claude Code run local servers on your own computer; those in Claude Desktop's config file are not available in claude.ai.

---

**Add Tracy's server and run one interview** Paste the server URL into Claude or ChatGPT and draft an interview in a chat. Approve the preview; the first 3 results are free. [Try it free](https://tracyinterview.ai/connect?utm_source=blog&utm_medium=seo&utm_campaign=what-is-a-remote-mcp-server&utm_content=try-free) [Book a demo](https://cal.com/tracy-ai/tracy-demo-web?utm_source=blog&utm_medium=seo&utm_campaign=what-is-a-remote-mcp-server&utm_content=demo)

## Sources

- [Model Context Protocol: Architecture overview](https://modelcontextprotocol.io/docs/learn/architecture), retrieved 14 Nov 2026.
- [Model Context Protocol: Authorization, version 2026-07-28](https://modelcontextprotocol.io/specification/2026-07-28/basic/authorization), retrieved 14 Nov 2026.
- [Claude Help Center: Get started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp), retrieved 14 Nov 2026.
- [Claude Code docs: Connect Claude Code to tools via MCP](https://code.claude.com/docs/en/mcp), retrieved 14 Nov 2026.
- [OpenAI docs: Connect and test your plugin in ChatGPT](https://developers.openai.com/plugins/deploy/connect-chatgpt), retrieved 14 Nov 2026.
- [OpenAI docs: Connect your MCP server to ChatGPT](https://developers.openai.com/api/docs/guides/developer-mode), retrieved 14 Nov 2026.
- [Tracy AI Interviewer: Add Tracy to Claude](https://tracyinterview.ai/help/claude), retrieved 14 Nov 2026.
- [Tracy AI Interviewer: Connect](https://tracyinterview.ai/connect), retrieved 14 Nov 2026.
- [Tracy AI Interviewer: Pricing](https://tracyinterview.ai/pricing), retrieved 14 Nov 2026.
