To configure a remote Model Context Protocol (MCP) server in ChatGPT, enable Developer mode, create a custom app, enter the server endpoint and metadata, authenticate if required, scan its tools, save a draft, test it in a new chat, and have a workspace administrator or owner publish it. ChatGPT connects to reachable remote servers; a server running only on your laptop or private network needs a supported connectivity path such as Secure MCP Tunnel.
The procedure below reflects OpenAI’s current Help Center guidance as of September 29, 2026. Plan eligibility, rollout status, permissions and interface labels can change, so verify the live Developer mode and MCP apps in ChatGPT article if your workspace looks different.
What “configure an MCP server from Chat” means
In this context, “Chat” means the ChatGPT web application, not an OpenAI API request. You are creating a custom ChatGPT app that points to an MCP server. The app advertises the server’s tools to ChatGPT, which can then call those tools during a conversation subject to permissions and approvals.
API integration is a separate path: an application sends an MCP tool definition containing a server URL or connector, authentication and approval settings in an API request. Do not paste API fields into the ChatGPT Apps screen. The API schema is documented in the OpenAI API Reference.
Recommended Free Tools
#1 Best Overall
Check eligibility before you start
- ChatGPT web: Custom apps and Developer mode are available for Business and Enterprise/Edu workspaces, subject to workspace controls and rollout.
- Business: Admins or owners control deployment. Pro users can connect MCPs with read/fetch permissions in Developer mode, while full support for write or modify actions is rolling out in beta to Business and Enterprise/Edu.
- Enterprise/Edu: An administrator may need to grant developer access through workspace role controls before a member can enable Developer mode.
- Authority: You need the server operator’s endpoint, required metadata and credentials, plus permission to expose the intended tools to your workspace.
If Developer mode or the creation controls are missing, the cause is usually plan, role, workspace policy or rollout—not a malformed server.
Prepare the MCP server
Use a reachable endpoint
ChatGPT connects to a remote MCP server. A service bound to localhost, a private LAN address or an on-premises network is not directly reachable from ChatGPT. Host the service at a permitted public endpoint, or use OpenAI’s Secure MCP Tunnel for a private, on-premises or developer-machine deployment. The tunnel avoids exposing the server directly to the public internet while providing a supported connection path.
Decide which tools to expose
Make a deliberate tool boundary before scanning. Read-only search, fetch or lookup tools are different from tools that create, delete, send or modify data. Grant only the actions users need, and document side effects, required parameters and expected authorization scopes. Search and fetch tools are no longer mandatory for connected servers.
Plan authentication and token renewal
Use the authentication method required by your server. If it uses OAuth or OpenID Connect, verify that the identity provider advertises the discovery metadata ChatGPT needs and issues refresh tokens. OpenAI’s guidance identifies offline_access as the standard scope to request a refresh token. Without a refresh-token scope or equivalent, access can expire after the initial authorization and users may have to authenticate again.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Rank #2
Configure the app in ChatGPT
- Enable Developer mode. An administrator first enables the feature in workspace settings where required. Depending on your plan and role, open workspace or user Settings, go to Apps or the Developer mode control, and turn it on. Labels differ between Business and Enterprise/Edu and may change.
- Open the creation screen. In the relevant Workspace or user Settings, choose Apps → Create. Confirm that Developer mode is enabled.
- Enter server details. Provide the MCP server endpoint and the required metadata, such as its name, description and any values requested by the form. Use the exact HTTPS endpoint supplied by the operator; do not add an MCP path or query parameter unless the operator specifies it.
- Select authentication. Choose the method supported by the server. Complete the OAuth authorization prompt if one appears. Confirm that the account being authorized has the least privilege needed for the tools you selected.
- Scan tools. Click Scan Tools and wait for the server’s tool definitions to load. Review names, descriptions, input fields and action types. A scan that succeeds does not prove that every downstream API or permission will work.
- Save a draft. Click Create. The app is now a draft available for testing, not necessarily for everyone in the workspace.
Test the draft safely
- Start a new chat and select the draft from the tools menu, or mention the app in your prompt.
- Begin with a harmless read-only request that exercises authentication and a representative input.
- Check the tool call, returned data, error messages and handling of missing or invalid parameters.
- Test authorization boundaries with an account that should not be able to access a protected record.
- For write or modify tools, use a sandbox or reversible test object first. ChatGPT may ask for confirmation depending on app permissions and context; some high-risk actions can be blocked.
- Repeat the test after OAuth tokens have been refreshed or renewed if your provider has short-lived access tokens.
App selection applies to the message that uses it. If a later turn needs fresh data or another action, select or mention the app again rather than assuming every subsequent message will call it.
Publish and manage the app
When testing is complete, an administrator or owner publishes the draft from Workspace settings → Apps → Drafts. Enterprise/Edu administrators can manage access and action permissions and refresh actions to review tool changes.
After approval, the workspace uses a frozen snapshot of the available tools and inputs until an administrator reviews and publishes updates. OpenAI’s current help guidance says Business apps could not be updated after publication and had to be recreated and republished; treat that plan-specific behavior as volatile and confirm it in the live documentation.
Before publication, decide who can discover the app, which roles can authorize it and whether write actions are allowed. Record the server version and tool list so a later refresh can be reviewed rather than accepted blindly.
Local and private MCP servers: what works
Can I connect to a local MCP server?
Not directly from ChatGPT. ChatGPT cannot reach a process that exists only on your computer or private network. Use a supported Secure MCP Tunnel or deploy the service at a remotely reachable endpoint. A tunnel or hosted endpoint must still enforce authentication, authorization and network controls.
Should I expose an internal server publicly?
Only if your organization’s security policy allows it and the endpoint is hardened. Prefer a tunnel for private or on-premises services when public exposure would create unnecessary risk. Whichever route you choose, restrict tools and credentials to the minimum required.
Security, privacy and approval controls
- Trust the operator: An MCP server can receive prompts, parameters and data needed to perform its tools. Unsafe servers can create security risks, including prompt injection.
- Review side effects: Identify tools that read, create, modify, send or delete information. Require confirmation for consequential actions where possible.
- Protect credentials: Use separate accounts or scopes for the app, rotate secrets and never place tokens in prompts or tool descriptions.
- Consider third-party retention: Remote MCP servers are third-party services. Data sent to them is subject to the operator’s retention and residency policies; review its terms before sending personal, confidential or regulated information. See OpenAI’s data-controls documentation.
- Audit changes: Compare tool definitions and input schemas whenever an administrator refreshes or republishes the app.
Common configuration failures and fixes
Developer mode or Create is missing
Check that you are using ChatGPT web, that your plan supports the feature, and that an administrator has enabled it or granted your role access. A workspace policy or staged rollout can hide the control.
Tool scan fails
Confirm the endpoint is reachable from the internet or through the supported tunnel, uses the exact URL supplied by the operator and presents a valid TLS certificate. Check server logs for rejected methods, malformed MCP responses, authentication failures and firewall blocks. Retry after correcting the server, then run Scan Tools again.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsRank #4
OAuth succeeds once, then stops working
Ask the identity-provider administrator to verify discovery metadata and refresh-token issuance. Request the offline_access scope when appropriate. If no refresh token is issued, reauthentication may be required whenever the access token expires.
A tool appears but returns unauthorized
The scan proves that the definition was readable, not that the connected identity can access every resource. Recheck scopes, tenant or workspace membership, resource-level permissions and any server-side allowlist.
Write action is blocked or asks for confirmation
That behavior can be intentional. Review app permissions, workspace action controls and the current beta rollout. Test with a lower-risk action and have an administrator approve the required capability.
The app works in testing but not for colleagues
The draft may not be published, access may be restricted by role, or users may need to select the app for their message. Ask an administrator to publish it from Workspace settings → Apps → Drafts and verify the intended audience.
Free tools Windows power users keep installed
One-click scans. No signup required.
Performance and operational checks
- Keep tool descriptions precise and input schemas narrow; this makes selection and validation more predictable.
- Set server-side timeouts and return actionable errors instead of a generic failure.
- Log request identifiers, authenticated principal, tool name and outcome without recording secrets or unnecessary personal data.
- Provide idempotency or safe retry behavior for operations that can be repeated after a network interruption.
- Monitor token expiry, tunnel connectivity, upstream API limits and changes to tool schemas.
API configuration is a different workflow
If you meant an OpenAI API integration rather than ChatGPT, define an MCP tool in your API request. Provide a server_url or connector_id (one is required), a server label and optional authorization, headers, allowed_tools and require_approval settings. Your application—not the ChatGPT Apps UI—runs the OAuth flow and supplies the access token. The reference lists connector identifiers for services including Dropbox, Gmail, Google Calendar, Google Drive, Microsoft Teams, Outlook Calendar, Outlook Email and SharePoint; validate the current schema before implementation.
Or skip the browser setup
If the MCP server you need is for website screenshots, ScreenshotNeo provides an MCP server that AI agents such as Claude, Cursor and other MCP clients can call. It removes cookie or consent banners, newsletter popups and chat widgets before capture; bot checks, blank pages and failed loads are not billed. It also reports the page verdict and billing status in response headers.
For a direct screenshot request, use the documented API at https://screenshotneo.com/docs/:
curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://stripe.com -o shot.webpThe same request in Python:
import requests
r = requests.get("https://api.screenshotneo.com/v1/shot", params={"access_key": "YOUR_API_KEY", "url": "https://stripe.com"}, timeout=90)
open("shot.webp", "wb").write(r.content)Or Node.js:
const q = new URLSearchParams({ access_key: 'YOUR_API_KEY', url: 'https://stripe.com' });
const res = await fetch(`https://api.screenshotneo.com/v1/shot?${q}`);ScreenshotNeo includes full-page and element capture, device presets, dark mode, retina scale, PDF output, custom CSS and JavaScript, waits, blocking controls, headers and cookies, geolocation, caching, signed links, asynchronous jobs, bulk capture and a usage API. The free plan includes 1,000 screenshots per month without a card; paid plans start at $5 for 3,000. Create a free ScreenshotNeo account.
Frequently Asked Questions
Are search and fetch tools required for connected servers?
No. OpenAI’s current Help Center guidance says connected servers no longer need to provide search and fetch tools.
Who can publish an MCP app in a workspace?
A workspace administrator or owner publishes the draft; Enterprise/Edu administrators also manage access and action permissions.
Does selecting an app persist for every message in a chat?
No. App selection applies to the message using it. Select or mention the app again when a later turn needs its tools.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
