Skip to main content
Feedback

OpenAI ChatGPT Setup and Configuration

OpenAI ChatGPT supports connecting custom remote MCP (Model Context Protocol) servers, allowing ChatGPT to access your company's private data, internal tools, and third-party systems directly within conversations. Custom MCP servers are registered as "apps" (formerly called "connectors") in ChatGPT and can be used in chat, deep research, and company knowledge workflows.

This feature requires Developer Mode and is available on ChatGPT Business, Enterprise, and Education plans with full write-capable MCP support. Plus and Pro individual users can connect, read/fetch-only custom MCP servers with Developer Mode enabled.

Prerequisites

Before connecting a custom MCP server to ChatGPT, ensure the following:

Your MCP server must be reachable over HTTPS. ChatGPT connects to remote MCP servers over the public internet. Servers behind VPNs or private networks will not connect. For local development, you can expose a local server using a tunneling tool such as ngrok or Cloudflare Tunnel.

Supported transport protocols. ChatGPT supports both SSE (Server-Sent Events) and Streamable HTTP MCP transports.

Supported authentication modes. ChatGPT supports OAuth, No Authentication, and Mixed Authentication (where initialize and list-tools calls are unauthenticated, while individual tool calls require OAuth or no auth based on per-tool security schemes).

Plan requirements. Full MCP support (including write actions) and Developer Mode are available on ChatGPT Business, Enterprise, and Education plans. Plus and Pro users can use Developer Mode for read-only custom MCP connectors.
Step 1: Enable Developer Mode

A workspace admin or owner must enable Developer Mode before anyone can create custom MCP apps.

  1. Navigate to Workspace Settings > Permissions & Roles > Connected Data.
  2. Find the Developer mode / Create custom MCP connectors toggle.
  3. Enable it.

For individual users on Plus or Pro plans, you can enable Developer Mode yourself:

  1. Go to Settings > Apps & Connectors > Advanced settings (at the bottom of the page).
  2. Toggle Developer Mode to ON.
  3. When prompted with the warning about unverified connectors, confirm to proceed.

Once Developer Mode is active, you will see a Create button under Settings > Apps & Connectors (or Settings > Connectors > Create depending on UI version).

Step 2: Create a Connector (App)

With Developer Mode enabled, create a connector for your custom MCP server:

  1. Ensure your MCP server is reachable over HTTPS. The URL should point to your server's /mcp endpoint (e.g. https://your-server.example.com/mcp).
  2. In ChatGPT, navigate to Settings > Connectors > Create (or Settings > Apps & Connectors, then click the Create button next to Advanced settings).
  3. Provide the required metadata:
    • Connector name — a user-facing title (e.g. "Internal CRM", "Project Tracker").
    • Description — explain what the connector does and when to use it. The model uses this text during tool discovery and selection, so be specific about what it can and cannot do.
    • Connector URL — the public HTTPS endpoint of your MCP server (e.g. https://abc123.ngrok.app/mcp).
  4. Pick an authentication mechanism if applicable (OAuth, No Authentication, or Mixed).
  5. Click Create.

The Chat GPT client has a couple of issues that may arise;

  1. Token endpoint auth method client_secret_post fails. Use client_secret_basic instead. Boomi Connect supports both.
  2. The first attempt to connect fails with Something went wrong with setting up the connection. Go into the app and click connect again and continue, this seems to go through an alternate path and should succeed.

If the connection succeeds, you will see a list of the tools your server advertises. If it fails, check the Troubleshooting section below.

After creation, the app appears as a draft:

  • In Workspace Settings > Apps > Drafts (for admins)
  • In your user settings under Settings > Apps > Enabled Apps with a "Dev" label

Step 3: Use the App in a Conversation

Once your connector is created, you can use it in ChatGPT:

  1. Open a new chat in ChatGPT.
  2. Click the + button near the message composer, then click More.
  3. Choose the connector for your app in the list of available tools. This adds your app to the conversation context.
  4. Prompt the model with queries related to your app's capabilities (e.g. "What are my open tasks?" for a project management app).

ChatGPT will display tool-call payloads in the UI so you can confirm inputs and outputs. Write tools will require manual confirmation unless you choose to "remember" the approval for the rest of that conversation. New conversations will prompt for confirmation again.

You can invoke multiple first-party and third-party apps in a single prompt — for example, retrieve internal data from one app and create a ticket in another based on the result.

Publishing the App

After testing, admins and owners can publish the app to make it available workspace-wide.

  1. Go to Workspace Settings > Apps.
  2. Click on Drafts, then click the Publish button.
  3. Review safety warnings (especially for write actions).
  4. Once published, the app appears in the workspace's approved connectors list and in users' Apps settings, labeled "custom".

Enterprise/Education plans have additional controls: admins can use RBAC to determine who gets developer mode access, who can access each published app, and which specific actions an app can take before publishing.

Business plans have a limitation: apps cannot be updated after publishing at launch. To change tools or metadata, you must recreate and republish.

Refreshing Metadata

When you change your MCP server's tools list or descriptions, you need to refresh the metadata in ChatGPT:

  1. Update your MCP server and redeploy it.
  2. In Settings > Connectors, click into your connector and choose Refresh.
  3. Verify the tool list updates and test with a few prompts.

New actions are not automatically available after a server update — you must use the Refresh button to examine and approve MCP server action changes

Other Clients

Once a connector is created in ChatGPT web, it is also available on:

  • ChatGPT mobile apps - the connector syncs automatically.
  • API Playground - visit the platform playground, open Tools > Add > MCP Server, and paste the same HTTPS endpoint. This is useful for viewing raw request/response logs.

Security Considerations

Custom MCP servers are not developed or verified by OpenAI. You are responsible for verifying that the server is safe and appropriate for your organization before publishing.

  • Only connect to trusted servers. A malicious server could exfiltrate sensitive data from anything that enters the model's context.
  • Be aware of prompt injection risks. Malicious MCP servers may embed hidden instructions in tool responses to manipulate ChatGPT's behavior. ChatGPT has built-in protections, but they are not absolute.
  • Review write actions carefully. ChatGPT displays explicit confirmation modals before write/modify operations. Only click "Allow" for tools you trust.
  • Use the "remember approval" option with care. Remembering approval applies for the rest of that conversation only. New conversations reset to requiring confirmation.
  • Inspect tool payloads. For each tool call, you can expand the details to view the full JSON inputs and outputs.

To report a malicious MCP server, email security@openai.com.

Troubleshooting

Connection fails when creating the connector: Verify the MCP server URL is correct, publicly reachable over HTTPS, and responds to the /mcp endpoint. If developing locally, confirm your tunnel (ngrok, Cloudflare Tunnel) is active.

OAuth flow not starting: Check that your authorization server's /.well-known/oauth-protected-resource and /.well-known/openid-configuration (or /.well-known/oauth-authorization-server) endpoints are reachable and returning correct metadata. Verify that offline_access is advertised in scopes_supported.

Tools not appearing: After creating the connector, click into it in Settings and use Refresh to reload the tool list. Verify your MCP server's tools/list response returns valid tool definitions.

Token expiry / re-authentication loops: Ensure your OAuth provider issues refresh tokens. If the provider does not advertise offline_access or refresh tokens are not being issued, enable this in your provider's admin console, then recreate the app so ChatGPT fetches the updated metadata.

App not available to workspace members: Only admins/owners can publish apps. Check that the app has been published from Workspace Settings > Apps > Drafts. For Enterprise/Education plans, verify RBAC settings allow the target users to access the app.

Write actions not working on Plus/Pro plans: Full write-capable MCP connectors are restricted to Business, Enterprise, and Education workspaces. Plus and Pro users are limited to read/fetch-only custom MCP connectors.

On this Page