Connectors in Claude and ChatGPT: Using MCP Without Writing Code

How to add custom connectors in Claude and use Developer mode in ChatGPT to work with MCP without code, plus safe permission settings.

You are not a programmer, but you keep hearing your team talk about MCP. The good news: you can use MCP without writing a single line of code. Both Claude and ChatGPT have settings pages where you add an MCP Server someone else has built, using just a URL. What you need to know is where to click and what to watch out for.

What is a connector?

A connector links an AI app to a remote MCP Server on the internet. The MCP guide Connect to remote MCP Servers describes Custom Connectors as the bridge between Claude and remote Servers, letting Claude use that Server's tools, resources and prompts. In ChatGPT, this capability lives under Apps and Developer mode.

Using connectors in Claude

According to the Claude Help Center, custom connectors are available on Free, Pro, Max, Team and Enterprise plans, with Free users limited to one.

Steps for Pro and Max

1. Go to Customize and choose Connectors
2. Click "+ Add", then "Add custom connector"
3. Enter a name and the remote MCP Server URL your provider gives you, review the login settings, and click Add
4. Complete the login steps, which are usually OAuth
5. In each chat, enable the connector from the "+" button, then Connectors

Team and Enterprise plans

An Owner adds connectors in Organization settings and chooses the authentication mode. Members then click Connect with their own accounts, so the organisation controls which Servers can be used.

Using MCP in ChatGPT

The OpenAI Developer mode guide explains that this mode gives ChatGPT full MCP support for both read and write tools, but it is labelled an elevated-risk feature.

1. Open Settings, go to Apps, then Advanced settings, and turn on Developer mode
2. Click "Create app" to register a remote MCP Server
3. On the app's details page, you can turn individual tools on or off
4. In a chat, choose Developer mode from the Plus menu and select the apps you want

Note that it works on the web only, and access differs by plan. The OpenAI Help Center says full MCP with write actions is in beta for Business, Enterprise and Edu plans.

Example: a freelance marketer

A freelance marketer uses a task management tool that offers an official remote MCP Server. She adds that URL as a custom connector in Claude, signs in with OAuth, and asks: "What tasks for Client A are due this week? Sort them by urgency." Claude pulls the task list and orders it. When she asks it to change a task's status, Claude asks for permission first every time.

Things to watch out for

• Anthropic does not verify custom connectors. Connect only to Servers from organisations you trust.
• Read the permissions at login and deny anything unnecessary.
• Use "Allow always" only for tools you trust to run unsupervised.
• Check inputs before confirming. OpenAI's guide warns that incorrect writes can destroy, alter or share data.
• Prompt injection. A dishonest Server can hide instructions. Turn off tools unrelated to the task at hand.
• Menus change often. Menu names and plan access may change, so always check the latest help pages.

You can use MCP without code, but read every permission prompt like you would read a contract.

How to get started

1. Check your vendor's documentation to see whether the tools you use daily offer an official remote MCP Server
2. Add one connector and try read-only tasks first
3. Keep tools that write or delete data turned off until you are comfortable
4. Once confident, enable writes, and confirm each action yourself

To understand the mechanics behind it, read Tools, Resources and Prompts, or try the course ลองใช้ MCP เชื่อม AI กับเครื่องมือจริง on Plants.work Learn.