# Best Way to Connect a Remote MCP Server to ChatGPT in 2026
> Connect ChatGPT to a remote MCP server with Developer Mode and OAuth, verify its tools, run a test prompt, and troubleshoot common connection failures.
- **Author**: parth-kanpariya
- **Published**: 2026-09-23
- **Modified**: 2026-09-23
- **Category**: AI & DevOps
- **URL**: https://kuberns.com/blogs/connect-remote-mcp-server-to-chatgpt/
---
To connect a remote MCP server to ChatGPT, enable Developer Mode, create a custom app or plugin, enter the server's remote HTTPS endpoint and choose its authentication method. ChatGPT then scans the endpoint for available tools and starts an authorization flow when access to an external account is required.
This guide demonstrates the complete process with the Kuberns remote MCP endpoint:
`https://mcp.kuberns.cloud/mcp`
After connecting and authorizing Kuberns, you can use the tools available to your account to inspect deployment status, build history, build and runtime logs, non-secret configuration, billing information and supported troubleshooting operations from a ChatGPT conversation.
This article is about connecting an existing remote MCP service to ChatGPT. If you created your own server and need to host it first, follow the separate guide to [deploy an MCP server to production](https://kuberns.com/blogs/deploy-mcp-server/).
## TL;DR: Connect a Remote MCP Server to ChatGPT
1. Open ChatGPT Apps or Plugins settings.
2. Enable Developer Mode.
3. Create a custom MCP app or plugin.
4. Enter the remote HTTPS MCP endpoint.
5. Select the required authentication method.
6. Complete authorization and review the requested permissions.
7. Wait for ChatGPT to scan the available tools.
8. Select the app in a new conversation and run a safe test prompt.
For Kuberns, enter `https://mcp.kuberns.cloud/mcp` as the Server URL and choose OAuth authentication.
## What Do You Need Before Connecting an MCP Server?
Before starting, confirm that you have:
- A remote MCP endpoint available over HTTPS.
- The authentication method required by the service.
- Login access to the account behind the MCP server.
- A supported ChatGPT plan and workspace configuration.
- Permission to enable Developer Mode or create a custom app.
- Access to ChatGPT through a supported web interface.
OpenAI's current Developer Mode and MCP apps documentation says that full MCP support is available for eligible Business and Enterprise/Edu workspaces on ChatGPT web. Pro users can connect MCP services with more limited read and fetch access in Developer Mode. Availability can also depend on the user's role and workspace administrator settings, so check the current documentation rather than relying on an older plan table.
You also need a trusted endpoint. A working URL does not prove that the tools behind it are safe. Confirm who operates the MCP server, what its tools can do and what account data the authorization screen requests before connecting it.
> **Nothing deployed on Kuberns yet?** [Deploy an application from your GitHub repository](https://dashboard.kuberns.com/) first, then return here to connect its supported deployment tools to ChatGPT.
## How to Connect a Remote MCP Server to ChatGPT
ChatGPT may display the connection under **Apps**, **Plugins** or a custom MCP option. Follow the labels available to your account while keeping the endpoint and authentication settings unchanged.
### Step 1: Open ChatGPT Apps or Plugins Settings
Open ChatGPT Settings and locate **Apps** or **Plugins**. In an eligible workspace, administrators may also create an app from **Workspace Settings → Apps → Create**.
Look for an option such as **Create**, **Create app** or **New Plugin**. The wording differs across current ChatGPT surfaces, but each path should open a form where you can provide the remote MCP endpoint and authentication method.
### Step 2: Enable Developer Mode
Turn on Developer Mode for the account that will create and test the connection. Depending on the workspace, the control may appear under **Settings → Apps → Advanced Settings** or in the workspace permissions for connected data.
If Developer Mode is missing, do not continue by guessing another setup path. Check whether the account has a supported plan, whether the workspace administrator has enabled custom MCP apps and whether your role is authorized to create them.
### Step 3: Create the Remote MCP Connection
Create a custom app or plugin and enter the connection details. For the Kuberns MCP service, use:
| Field | Value |
|---|---|
| Name | Kuberns MCP |
| Description | Access supported Kuberns deployment tools |
| Connection | Server URL |
| Server URL | `https://mcp.kuberns.cloud/mcp` |
| Authentication | OAuth |
Review the URL carefully before proceeding. The hostname should be `mcp.kuberns.cloud`, the connection should use HTTPS and the path should end in `/mcp`.
The interface may display a warning that custom MCP servers can introduce risk. Read it before confirming that you understand and want to continue. This warning is relevant because an MCP tool may read external information or perform a supported action using the permissions granted to the connected account.
### Step 4: Sign In and Approve Access
After you create the connection, ChatGPT contacts the endpoint and starts the OAuth flow. Select **Sign in with Kuberns** and authenticate with the Kuberns account containing the applications you want to inspect.
The Kuberns authorization screen used for this guide requests permission to:
- View deployment status and build history.
- View build and runtime logs.
- View environment and resource configuration without secret values.
- View billing and credits information.
- Create and update support tickets.
- Debug the application.
Review the current consent screen rather than approving it automatically. The displayed permissions are the clearest description of the access being requested. In the tested flow, the environment-configuration permission explicitly excludes secret values.
Select **Approve** only when the requested access matches what you intend to use. OAuth connects ChatGPT to the selected Kuberns account, but it does not make every tool or prompt inherently safe.
### Step 5: Confirm the MCP Tools Are Available
Wait for ChatGPT to finish scanning the remote endpoint. Then open the Kuberns app or plugin and confirm that its tools appear with names, descriptions and input requirements.
The verification should establish that:
- Kuberns appears in enabled Apps or Plugins.
- The tool scan completed without an authentication error.
- Deployment-related tools are visible.
- The visible tools match the permissions you approved.
- Write or destructive actions are clearly labelled when applicable.
Tool definitions can change. OpenAI notes that updates to an MCP server's tools are not always enabled automatically. Depending on the workspace, an administrator may need to refresh the available actions or recreate and republish the app.
### Step 6: Run a Safe Test Request
Open a new supported conversation and select Kuberns from the tools menu. If the interface supports app mentions, you can also mention Kuberns directly in the prompt.
Start with a read-only request:
> Use Kuberns to show the status and build history of my most recently deployed application.
A successful response should prove that ChatGPT recognized the Kuberns app, the authorization succeeded, an available tool was invoked and the returned information belongs to the connected account.
The current OpenAI MCP app guide explains that app selection can apply to the message where it is used rather than the entire conversation. Select or mention Kuberns again when a later message needs fresh deployment data or another tool call.
## What Can You Do With the Kuberns MCP Server?
The useful actions depend on the tools currently exposed by Kuberns and the permissions granted to the connected account. The following requests match the capabilities shown in the tested authorization flow.
| Task | Example prompt |
|---|---|
| Check deployment status | “Use Kuberns to check the status of my latest deployment.” |
| Review build history | “Summarize the last three builds for my API service.” |
| Diagnose a failed build | “Review my latest failed build and identify the first meaningful error.” |
| Inspect runtime logs | “Find recurring errors in the recent runtime logs for my backend.” |
| Review configuration | “List the configured environment variable names without showing secret values.” |
| Check resources | “Show the resource configuration for this application.” |
| Review billing information | “Show my current Kuberns credits and billing information.” |
| Contact support | “Create a support ticket using the relevant details from this failed build.” |
Use an explicit request such as “Use Kuberns” when ChatGPT answers generally instead of invoking the connected app. Include the application or service name when the account contains several projects.
For a broader workflow across several services, read how to [manage multiple application deployments](https://kuberns.com/blogs/manage-multiple-deployments/). If a tool identifies a build error but you need more context, review the [common reasons software deployments fail](https://kuberns.com/blogs/why-do-software-deployments-fail/).
## Can ChatGPT Deploy an Application Through an MCP Server?
ChatGPT can perform only the actions exposed by the connected MCP server and permitted by the user's authorization, workspace controls and confirmation settings. Adding an MCP endpoint does not automatically allow ChatGPT to deploy, restart, scale, roll back or modify an application.
Before using any write operation, check the current tool name, description, input fields and risk labels. ChatGPT may request confirmation based on the permission and context, and some high-risk actions may be blocked. Do not assume a requested action succeeded until the tool returns a clear result and the corresponding state is verified in the provider.
The Kuberns flow verified for this guide supports visibility into deployment status, build history, build and runtime logs, non-secret environment and resource configuration, billing and credits, support tickets and application debugging. Any additional deployment action should be documented only after its corresponding production tool has been tested.
## Why Use the Kuberns Remote MCP Server With ChatGPT?
ChatGPT provides the conversational interface, while Kuberns supplies the authorized deployment context and supported operational tools. MCP standardizes the connection between them, and OAuth ensures the requests use the Kuberns account approved by the user.
This means a developer can ask a deployment question without first copying a status page, log output and configuration details into the conversation. The connected tools can retrieve the supported context directly, subject to the permissions and confirmation controls shown to the user.
[Kuberns](https://kuberns.com/) is an Agentic AI platform for deployment designed for full-stack and complex backend projects. Its remote MCP service gives ChatGPT controlled access to supported deployment and troubleshooting operations while Kuberns remains the platform behind the running application. New users can read more about [how Kuberns deploys and manages applications](https://kuberns.com/blogs/what-is-kuberns-the-simplest-way-to-build-deploy-and-scale-full-stack-apps/).
> Bring the deployment context into the ChatGPT workflow you already use for development. [Deploy your application on Kuberns](https://dashboard.kuberns.com/) and connect the remote MCP endpoint when it is ready.
## Is It Safe to Connect a Remote MCP Server to ChatGPT?
A remote MCP connection can be useful, but it should be treated as access to an external service rather than a harmless browser setting. Connect only endpoints operated by providers you trust and review the permissions before enabling their tools.
The OpenAI safety guidance for MCP apps warns that unsafe or untrusted servers can introduce risks such as prompt injection and unintended data access. It also explains that confirmation requirements vary with the action, permission and context.
Use these safeguards:
- Confirm the endpoint's domain before connecting.
- Approve only the permissions required for the task.
- Treat tools labelled write, open-world or destructive with additional care.
- Read every confirmation request before accepting it.
- Avoid including unnecessary customer data in prompts.
- Do not paste secret values into the conversation.
- Disconnect access when it is no longer required.
- Follow a documented process to [manage production environment variables safely](https://kuberns.com/blogs/environment-variables-in-production/).
The Kuberns consent screen shown in this guide states that environment and resource configuration is visible without secret values. Recheck that wording whenever reconnecting because permissions can evolve with the tool inventory.
## Common Remote MCP Server Connection Problems
The connection process can fail even when the endpoint itself is online. Use the visible error and the stage where the failure occurred to narrow down the cause.
| Problem | Likely cause | What to do |
|---|---|---|
| Developer Mode is missing | Plan, role or workspace restriction | Check current eligibility and ask the workspace administrator to enable custom MCP access |
| Create App or New Plugin is unavailable | The account cannot create custom connections | Use an authorized administrator or developer account |
| ChatGPT rejects the MCP URL | The endpoint is local, invalid, unreachable or not using the expected protocol | Confirm the remote HTTPS URL and test the provider's documented endpoint |
| OAuth does not begin | The wrong authentication method was selected or discovery failed | Select OAuth, confirm the endpoint and recreate the connection |
| Authorization completes but no tools appear | The scan did not finish or authorization was not retained | Reopen the connection, reconnect the account and repeat the tool scan |
| The app is connected but unavailable in chat | The app is not enabled for the current message or the chat surface is unsupported | Open a new supported web chat and select or mention the app |
| ChatGPT answers without using the MCP tool | The prompt is ambiguous or the app was not selected | Explicitly say “Use Kuberns” and name the deployment task |
| ChatGPT shows an old tool list | The server changed after the app was created | Refresh its actions or recreate the app according to the current workspace controls |
| The connection repeatedly expires | OAuth refresh access is unavailable or authorization was revoked | Reconnect the account and complete the authorization flow again |
| A local STDIO configuration does not work | This setup expects a remote MCP endpoint | Use the provider's remote HTTPS endpoint or an officially supported secure tunnel |
These problems also appear in community discussions. One developer asked where a custom ChatGPT MCP app had gone and how to refresh its tools. Another discussion describes MCP connections visible in settings but unavailable in conversations. These threads demonstrate the user problems, while OpenAI's current documentation should remain the authority for the solution.
## How to Disconnect a Remote MCP Server From ChatGPT
Open ChatGPT Settings, select **Apps** or **Plugins**, and open the connected MCP app. Use the available disconnect, disable or remove control. If the service provides its own authorization-management page, revoke the ChatGPT connection there as well.
Workspace-managed apps may need to be disabled by an administrator. Changing whether ChatGPT asks for confirmation is not the same as revoking access. The OpenAI connected apps documentation explains that users can disconnect apps from Settings, while workspace administrators can disable centrally managed access.
Reconnect the app when authorization expires or when you intentionally need its tools again. Review the current permissions as if it were a new connection.
## Connect Your Remote MCP Server to ChatGPT
Connecting a remote MCP server to ChatGPT requires a reachable HTTPS endpoint, the correct authentication method, a successful tool scan and explicit user authorization. Once the Kuberns MCP service is connected, developers can bring supported deployment status, logs, configuration, debugging, billing and support operations into the same ChatGPT workflow they already use for development.
Deploy your application on Kuberns, connect `https://mcp.kuberns.cloud/mcp` to ChatGPT and begin with a read-only deployment-status request. Expand to other tools only after reviewing their permissions and expected effects.
[](https://dashboard.kuberns.com/)
## Frequently Asked Questions
### How do I connect a remote MCP server to ChatGPT?
Enable Developer Mode in an eligible ChatGPT account, create a custom app or plugin, enter the remote HTTPS MCP endpoint, choose the required authentication method and complete the tool scan. If authorization is required, sign in to the provider and approve the displayed permissions before testing the app in a new conversation.
### What URL should I use for the Kuberns MCP server?
Use `https://mcp.kuberns.cloud/mcp` as the Server URL when creating the Kuberns custom MCP connection in ChatGPT. Select OAuth as the authentication method and complete the Kuberns authorization flow.
### Can ChatGPT connect directly to a local MCP server?
The custom connection described in this guide uses a remote HTTPS MCP endpoint. ChatGPT does not directly launch a local STDIO server through this flow. OpenAI documents Secure MCP Tunnel as an option for supported private, on-premises or developer-machine servers.
### Why can’t I find Developer Mode in ChatGPT?
Developer Mode availability depends on the ChatGPT plan, workspace, account role and administrator settings. Check the current OpenAI eligibility documentation and ask the workspace administrator to enable custom MCP access when the setting is restricted.
### Why is my MCP server connected but its tools are missing?
The tool scan may not have completed, authorization may have expired, the app may not be enabled for the current chat or ChatGPT may have an older tool snapshot. Reopen the connection, complete authorization, refresh its actions when available and test it in a new supported conversation.
### Can ChatGPT see Kuberns environment secret values?
The Kuberns authorization screen used for this guide states that ChatGPT can view environment and resource configuration without secret values. Review the current consent screen before approving because the displayed permissions are the authoritative description of the access being requested.
### Can ChatGPT deploy or modify my Kuberns application?
ChatGPT can perform only the actions exposed by the connected Kuberns MCP tools and allowed by the user's authorization and workspace controls. A connection alone does not grant every deployment or infrastructure action. Review the current tool list and confirmation request before approving any write operation.
### How do I disconnect the Kuberns MCP server from ChatGPT?
Open ChatGPT Settings, select Apps or Plugins, open the Kuberns connection and choose the available disconnect or disable option. If Kuberns provides a separate authorization-revocation control, revoke access there as well. Workspace-managed apps may need to be disabled by an administrator.
---
- [More AI & DevOps articles](https://kuberns.com/blogs/category/ai-devops/1/)
- [All articles](https://kuberns.com/blogs/)