Using MCP Without Writing Code

You can use the Model Context Protocol without writing a line of code. If the service you want to reach runs a remote MCP server, all you need is its web address, an assistant that accepts custom connectors (Claude and ChatGPT both do, on some plans), and your own login for the service. You paste the address into the assistant’s settings, sign in through your browser, choose what the assistant may do, and start asking. Nothing is installed and no configuration file is edited. If you want to know what MCP is before connecting anything, start with the glossary entry.

What you need before you start

• The server’s address. It is a web address, usually ending in /mcp, and the service’s help pages will give it. If the service only offers something to “run with npx” or “add to a config file”, that is a local server, and it needs the developer route.
• An account on the service, with the access you actually want the assistant to have. The assistant never gets more than your login allows.
• An assistant plan that allows custom connectors. The option sits in different places on different plans, and on some it is switched off by an administrator.
• Ten minutes, and a few real items in the service so the first answer is about something you recognise.

Adding a server in Claude

Anthropic’s guide to custom connectors using remote MCP↗ (opens in a new tab) gives the steps for Pro and Max accounts, and they are all clicks.

1. Open Customize, then Connectors.
2. Press the plus button, then Add custom connector.
3. Paste the server address. Leave Advanced settings alone unless the service gave you an OAuth client ID and secret to enter there.
4. Press Add. Claude sends you to the service to sign in and approve.
5. In a chat, open the plus menu at the bottom left, choose Connectors, and switch the new one on for that conversation.

On Team and Enterprise plans, an owner adds the connector first under Organization settings, Connectors, and members then connect their own accounts to it. Many well-known services are also listed in Claude’s Connectors Directory↗ (opens in a new tab), where you connect with one button and no address at all; each listing says what the connector can read and write. Anthropic’s help pages say custom connectors are open to Free, Pro, Max, Team and Enterprise users, with Free limited to one.

Adding a server in ChatGPT

ChatGPT reaches arbitrary MCP servers through developer mode. OpenAI’s developer mode guide↗ (opens in a new tab) says it gives full MCP client support for all tools, read and write, and is available to Pro, Plus, Business, Enterprise and Education accounts on the web. Despite the name, turning it on is a switch, not a coding task.

1. Open Settings, then Security and login, and turn on Developer mode.
2. Go to Plugins and press the plus button to create an app.
3. Enter the server address, including the /mcp part, and choose the sign-in the service uses (usually OAuth).
4. Sign in when ChatGPT sends you to the service, then review the list of tools it found before you save.

OpenAI’s page on connecting a server to ChatGPT↗ (opens in a new tab) walks through the same screens. In a Business or Enterprise workspace, administrators may control whether members can do this themselves. Menu names in ChatGPT move often; if a label here does not match your screen, search your settings for developer mode.

Reading the sign-in screen

The step that matters most is the one people click through fastest. Remote MCP servers sign assistants in with OAuth, the same kind of “allow this app” screen you see when a calendar app asks for your Google account; the MCP authorisation specification↗ (opens in a new tab) builds on it. The screen is shown by the service, not by the assistant, and it should tell you four things.

• Which app is asking. The name is chosen by the app, so check it matches what you started a moment ago. If you did not just start a connection, press Cancel.
• Which account you are signed in as, and which workspace or board the connection will work in. Assistants act as you, so this is the identity everything will be recorded under.
• What it may do. Good servers let you choose: read only, or read and change, or comment. Tick the least you need today.
• Where to take it back. Somewhere in the service’s settings there should be a list of connected assistants with a revoke button.

On fenbs, for example, the approval page names the app and says plainly that the name is unverified, shows the board it will work in and the email you are signed in with, and offers three ticks: read the board (always on), add and change tasks, and comment. The connection then appears by name under Settings, “Connect an AI assistant”, where revoking it stops it at once. How MCP sign-in works covers what happens behind the screen.

Two layers of permission

There are two separate controls, and it helps to know which is which. The service decides what is possible: your role there, narrowed by the ticks on the sign-in screen. The assistant decides when to stop and ask you: Claude asks for approval before using a tool until you choose to always allow it, and ChatGPT’s developer mode asks for confirmation before write actions by default. The second layer never grants anything the first refused; it only adds a pause.

Keep the pause on for anything that changes data. Allow reads freely once you have seen a few, and read the details of each write before you approve it: which item, which field, what new value.

Staying safe when you are not a developer

• Only connect servers from the service itself or from people you trust. A malicious server can hide instructions in what it sends back, the attack known as prompt injection; Anthropic and OpenAI both warn about it.
• Start read-only. A week of questions tells you how the assistant understands the service before it can change anything.
• Switch connectors on per chat. A connector that is off cannot be steered by a stray instruction in a document you pasted.
• Never paste passwords or tokens into a chat. A proper remote server never asks for them; the browser sign-in replaces them.
• Check what it did. Services with an activity log or history show every change under a name. Look after the first few writes.
• Disconnect what you stopped using. Revoking takes a minute and removes a door you are no longer watching.

A first hour that teaches you something

1. Ask who it is: “Which account are you connected as, and what can you do?” Many servers have a tool that answers exactly that.
2. Ask for a list: “What is on my board in Next Up?” Check it against the service.
3. Ask for a summary: “What changed this week, by person?”
4. Make one small change with the pause on: “Add a task called ‘Try the connector’ to To Do.” Read the approval, allow it, and find the task in the service.
5. Revoke and reconnect once, so you know where the switch is before you need it.