Phase 1 · ChatGPT · Level 3 · Power User
Connectors to your work apps and MCP
By the end, you'll be able to…
- Explain what a connector does and set expectations for connecting a work app like Drive or SharePoint
- Use ChatGPT against your own connected files instead of uploading them each time
- Understand MCP in plain terms (the common standard connectors are built on) and the trust it demands
Why it matters
Uploading the same document every time you want ChatGPT to work with it gets old fast. Connectors let ChatGPT reach into apps you already use (Google Drive, SharePoint and others) so it can draw on your real files and data with your permission, no re-uploading. And behind many connectors sits MCP, an emerging common standard that's worth understanding in plain English, because it's why this ecosystem is growing so quickly, and why 'only connect what you trust' is the rule that matters.
What a connector is
A connector links ChatGPT to another system you use (your files, your email, a document store) so ChatGPT can read from (and sometimes act in) that system with your permission. It's what lets the tool step outside the chat window and work with your actual material instead of whatever you happened to paste in. Connect your Google Drive, and you can ask ChatGPT about a document that lives there without downloading and uploading it first. The file stays where it is; ChatGPT reaches it through the connector when you point it at one.
You'll find and manage connectors in ChatGPT's settings, usually under an Apps or Connectors area, a connector directory that lists what you can turn on. The catalogue has grown quickly: Google Drive, SharePoint, Dropbox, Box and more, with additional apps available for research-style tasks. Connecting one typically takes a couple of minutes: pick the app, you're sent to that service's own sign-in and permission page, you grant ChatGPT access, and you're back. That redirect-and-grant step is the service asking you to authorise the link; it's not ChatGPT quietly reaching in.
Two caveats to set expectations honestly. First, availability varies by plan and, just as much, by region and by your organisation's settings: some connectors are limited to business or enterprise plans, and what's on offer differs between countries and can change, so what you see in the directory is the reliable guide, not any fixed list. Second, if you're on a managed work account, your IT team may control which connectors are allowed; a connector that's greyed out isn't broken, it's governed.
Using ChatGPT against your connected files
Once a connector is set up, the shift in day-to-day use is small but real: you stop uploading and start pointing. Instead of "here's a document I've attached", it's "the document called X in my Drive". The workplace payoff is for the material you return to repeatedly (the policy, the brief, the running spreadsheet) which now lives one reference away instead of one upload away.
Using my connected Google Drive, find the document called "Fernway feedback process brief" and summarise its success measures and timeline in a short list. First, confirm the exact file name and its last-modified date so I know you've opened the right one; if there's more than one document with a similar name, list them and ask which I mean before summarising.
Why this works: Naming the file precisely and asking ChatGPT to confirm which document it opened turns a vague 'my Drive' request into a checkable one. Connectors can pull the wrong file when names are similar, so confirming the source before you act on the answer is the habit that keeps this reliable.
The everyday habit here mirrors the one from the Projects lesson: confirm the source. A connector makes many files reachable, and similar names or old versions are easy to grab by mistake. Asking "which file did you use, and when was it last changed?" costs a sentence and saves you acting on the wrong document.
Using my connected Drive, look at both the "Fernway project brief" and the "operations sync notes" from this month, and tell me: which actions in the notes relate to the project, and which are still unassigned. For each point, say which document it came from so I can verify it. If something isn't covered in either file, say so rather than filling the gap.
Why this works: Connectors earn their keep when a question spans several documents you'd otherwise open one by one. Being explicit about which sources to use, and asking it to say where each point came from, keeps the answer traceable instead of an unattributed blend you can't check.
MCP, in plain English
Behind a lot of this sits a piece of plumbing worth understanding, even though you'll rarely touch it directly: MCP, the Model Context Protocol. In plain terms, MCP is a common plug. Before it, connecting an AI tool to each new service meant custom work for that one service, a different adaptor every time. MCP is an open standard that lets any tool which "speaks MCP" connect to any service that also speaks it, without a bespoke build for each pairing. It's the USB-C of AI connections: one shape, many devices.
Why should a non-developer care? Two reasons. First, it explains the pace: the reason the connector directory keeps growing is that services can plug in through a shared standard rather than waiting for a one-off integration. Second, and more practically, ChatGPT can connect to custom MCP tools your organisation might build or adopt (an internal system, a niche database) through this same standard. So "can ChatGPT work with our in-house tool?" increasingly has the answer "if someone exposes it over MCP, yes." Setting those up is a developer job (it lives behind a developer mode and is aimed at organisations), but knowing the concept means you understand what's possible and can ask the right question of your IT team.
The flip side of a common plug is a common risk, and it's the same instinct that runs through this whole level. An MCP connection, especially a custom or third-party one, is a channel through which ChatGPT reads external content and can sometimes take actions. That makes it a route for prompt injection: a malicious server or a poisoned document could carry hidden instructions. The rule is blunt and worth memorising: only connect tools, accounts and servers you actually trust and are permitted to link. A connector you don't understand is not a convenience; it's an unknown door.
List the connectors and apps I currently have connected, and for each one, say in plain English what it lets you access: read-only, or can you also make changes? I want to check nothing is connected more broadly than I intended.
Why this works: Before leaning on a connection, it's worth making its scope explicit. Asking ChatGPT to state what a connector gives it access to helps you notice if a connection is broader than you intended; a five-second check that supports the 'only connect what you trust' rule with 'and know what you've connected'.
Try it now
Common mistakes
- Connecting broadly and forgetting. Every connector is a standing grant of access to real data. Connect what you'll use, know what each one can see, and disconnect what you don't need; don't accumulate open doors.
- Ignoring the managed-account rules. On a work account, connectors touch organisational data and may be governed for good reason. Check policy before connecting; a governed or greyed-out connector is a decision, not a fault.
- Assuming it grabbed the right file. Similar names and stale versions make wrong-file answers easy and invisible. Ask which document and which version it used before acting on the result.
- Trusting a connected-file answer more because it came from "your own" data (over-trust). An answer sourced from your real Drive feels solid precisely because it's your material, but ChatGPT can still open the wrong version, misread a table, or blend two documents into a claim neither makes, and the connector's authority makes that error more convincing, not less. Have it name the source and quote the relevant line, and check anything load-bearing against the file itself. And treat a connection to anything you don't fully trust as a genuine security risk, not just a convenience; the common plug carries the common danger.
Keeping current
Which connectors exist, which plans and regions get them, how custom MCP tools are enabled, and the exact safeguards are all changing quickly. For the current picture, see OpenAI's help articles on Apps and connectors in ChatGPT and Developer mode and MCP apps, plus the ChatGPT release notes. Accurate as of 13 July 2026.