A Project's knowledge base (lesson 5.1) holds a snapshot: whatever you uploaded, as of when you uploaded it. A connector is different — it gives Claude live access to an external data source, so a conversation can query the current state of Drive, Gmail, or another connected service directly, rather than working from a copy that can drift out of date. If a contract template in your knowledge base gets superseded next quarter, the knowledge base still has the old version until someone re-uploads it; a Drive connector pointed at that same template's live file reflects the update the moment it happens.
This distinction matters for configuration decisions: material that's genuinely stable and worth having reviewed once (a finalized style guide) is a reasonable knowledge base upload. Material that changes on its own schedule outside the Project — a shared planning doc, an inbox, a live spreadsheet — is a better fit for a connector, so the Project is always working from the current version instead of a copy someone has to remember to refresh.
Common Connectors: Drive and Gmail
The Google Drive connector lets Claude search and read files in a connected Drive directly from a conversation — pulling in the current content of a doc or spreadsheet without anyone manually opening it, copying text, and pasting it in. The Gmail connector works the same way for a connected inbox: Claude can search and read messages and threads to ground a task in the actual current state of a conversation or a specific email, instead of you summarizing the thread yourself first. Both are examples of the same underlying idea — a connector as a bridge from a Project or conversation into a live external service, configured once (by connecting and authorizing the service) and then available going forward without repeating that setup on every task.
Reading and Writing Back
It's a natural assumption that a connector is read-only — that Claude can look things up in Drive but you'd still have to manually download and re-upload anything Claude generates. For the Drive connector, that's not the limit of what's supported: Claude can create folders and save Claude-generated files directly into Drive itself, not just read existing files from it. If you ask Claude to draft a report and save it into a specific Drive folder, the Drive connector can fulfil that directly — creating the destination folder if it doesn't already exist, and saving the generated file into it — without you exporting and re-uploading the content by hand. Different connectors support different combinations of read and write access depending on the underlying service and how it's configured, so it's worth checking what a specific connector actually supports rather than assuming every connector behaves identically to every other one.
Key Concept
A connector gives Claude live access to an external data source, in contrast to a knowledge base upload, which is a fixed snapshot as of upload time. The Drive connector in particular supports writing back — creating folders and saving Claude-generated files directly into Drive, not just reading from it.
Common Exam Distractor
Watch for answers assuming a connector is strictly read-only, or that saving a generated file into Drive still requires a manual download-then-upload round trip — that manual step is exactly what a write-capable connector like Drive is built to avoid. Also watch for treating a connector and a knowledge base upload as interchangeable: a connector reflects the source's current state, an upload reflects only what it looked like at upload time.