How to Give ChatGPT Permission to Use Connected Tools

ChatGPT gets permission to use outside tools through connections you set up in your account settings. Each connection goes through an authorization step where you approve what ChatGPT can do in that service. Tools like fenbs let you record the approved task-level boundaries for each connection so they are clear, attributable, and available to any AI app working on the same project.

This guide explains how ChatGPT permissions work, where to review and manage connected tools, what each permission level means, and how to keep a clear record of what you have authorized.

How ChatGPT app permissions work

When you connect a service to ChatGPT, such as Google Drive, GitHub, or a calendar, the service asks you to approve a set of permissions through an OAuth authorization flow. These permissions define what ChatGPT can read, write, or do in that service on your behalf.

ChatGPT uses those permissions whenever you ask it to access that service in a conversation. It cannot do more than what you approved, and approving access does not mean ChatGPT will use it constantly. It calls the connection only when you ask it to.

Where to review connected apps

In ChatGPT, go to your account settings and look for connected apps or integrations. This shows which services have been authorized and what level of access each one has. The same information is available in each connected service: check your Google Account, GitHub account, or other service under Security or Authorized Applications.

Review this list periodically. Connections you no longer use are worth removing to limit unnecessary access. A service you authorized months ago may have a broader scope than a current task needs, and removing it reduces the surface area if something goes wrong.

How to add a new permission or connection

To add a new connection, find the service in ChatGPT's integrations panel and click to connect. You are taken through the service's OAuth sign-in, shown the permission scope being requested, and asked to approve or deny it. Read the permission scope before approving.

If the scope seems broader than what you need, check whether a more limited option is available. For Google Drive, "read-only access to specific files" is safer for most tasks than "edit and delete all files." Approve only what the task actually requires.

What each permission level allows

OAuth scopes describe what ChatGPT can technically do in a service. Read scopes allow viewing content without changing it. Write scopes allow creating or editing. Delete scopes allow removing content. A scope labeled "full access" usually combines all three.

The scope is a technical permission, not a task description. "Edit all files in Google Drive" is the permission. "Review the project folder documents and suggest changes in the chat" is the task. The gap between those two descriptions is where unintended actions can happen. The safest approach is to pair the OAuth connection with a written note that describes what the AI may do, what it should ask about before doing, and what is outside its scope. That note is a task-level agreement rather than a technical constraint, but it is what prevents the gap from becoming a problem.

How to revoke access

Revoke a connection from two places: ChatGPT's connected apps settings, and the connected service's own security page. Both work. Revoking from the service's security page is the more authoritative option: it removes the OAuth token directly and takes effect immediately.

When you revoke access, ChatGPT can no longer use that service in future conversations. Any new request to access the service will fail until you reconnect. Revoking does not delete data that was already created or accessed during earlier sessions.

Keeping approved task-level boundaries clear

The OAuth scope tells you what is technically permitted, but it does not describe the task-level rules your team has agreed on. "ChatGPT may read Google Drive" is the permission. "ChatGPT may read the project folder but must ask before opening anything in the finance folder" is the decision. One is stored in an account setting; the other needs to be written somewhere people and AI apps can read it.

Recording task-level rules in fenbs means they are available at the start of every session, visible to a teammate, and readable by other connected AI apps. A rule entered in fenbs appears in the AI context before the assistant does any work.

How to do this with fenbs

1. Create a free account. Your personal board is free and needs no credit card.
2. Add a project for each workspace where AI tools have connected access.
3. Write the approved task scope in Decisions and rules: what each AI app may do, what it must ask a person about, and what is off-limits.
4. Connect ChatGPT or another supported app over MCP from Integrations. Those rules appear in the AI context automatically.
5. Review and update the rules when you add a new connection, change the scope of a task, or revoke an existing connection.

fenbs does not manage OAuth permissions for other services. It keeps the agreed task-level rules that go beyond what an OAuth scope can describe.