Claude, Google Drive API

A common assumption is that Claude finds a file in Google Drive by walking its folder path, the way a normal file system works. The video corrects that: there is no path API, and Drive works entirely by file ID. It then traces one realistic request through the API to show three habits that decide whether Claude's calls come back complete or quietly wrong.

Drive has no path API

Everything in Drive is addressed by ID. A file does not live at a path such as a folder name followed by a file name, so Claude cannot ask Drive to resolve one. Any plan that depends on walking a path has nothing to call, which is why the ID-based model shapes every request Claude makes.

A folder is just a file with a mime type

In Drive, a folder is not a separate kind of thing from a file. It is a file whose mime type says it is a folder. Hierarchy exists only through each file's parents array, which holds the IDs of the items that contain it.

That has a practical consequence. There is no "get folder" endpoint that returns a folder's contents. When Claude is asked to list what is in a folder, it searches for every file whose parents array contains that folder's ID. The folder's contents are a filter result, not a lookup.

Shared drives are invisible by default

The video's worked example is a request to export every spreadsheet modified in the last month on a team's shared drive as CSV. Files on a shared drive are invisible by default, so the search needs three extra parameters before it can see them at all. The video names supportsAllDrives as one of them, and points out that leaving it out does not produce an error. The file simply looks as if it was never there.

Pagination must be requested

Once the search is set up correctly, the response can still be incomplete. The fields= parameter has to explicitly ask for nextPageToken, otherwise the results quietly stop at one page. With the token requested, the list comes back complete, page after page.

Google-made files cannot be downloaded

The last trap is the file type. A Google Sheet has no downloadable bytes, because it is not a binary file. Pulling it with alt=media returns a flat 403 rather than a helpful message. The fix is to use the export endpoint instead, which is how the example spreadsheet becomes a CSV.

Failures that never name the cause

The video stresses that skipping any one of these habits produces a failure that does not say which parameter was missed. A missing supportsAllDrives just looks like the file was never there, and the 403 on a Sheet is a flat refusal rather than a helpful error. That is why the video treats these as habits to apply on every Drive call rather than errors to debug afterward.

Key takeaways

• Drive has no path API; everything works by file ID.
• A folder is a file with a folder mime type, and hierarchy lives in each file's parents array.
• Listing a folder means searching for files whose parents contain the folder's ID.
• Files on a shared drive are hidden unless the search includes the three extra shared-drive parameters, including supportsAllDrives.
• Ask for nextPageToken in fields= or results stop at the first page.
• Google Sheets cannot be fetched with alt=media; use the export endpoint instead.
• These failures do not name the missing parameter, so check all three on every call.

Try it yourself

Give Claude this prompt: "Find all spreadsheets modified in the last 30 days on our shared drive, and export the first one as CSV." Then watch three things. Does it pass all three shared-drive parameters on the search? Does it request nextPageToken in fields= instead of assuming one page is everything? Does it reach for the export endpoint instead of alt=media once it sees the result is a Google Sheet?