google-drive

# Google Drive files ## Ids come from search, never guesses Every tool takes Drive file ids. Get them from `search` (or `list_folder` for a known folder, `list_recent_files` for "what was I just working on"), or from the id in a URL the user pasted. `exact_name` beats `query` for a known filename; add `mime_type_filter` to cut noise. Never invent an id or reuse one from a stale conversation - re-search first. ## Folder targeting - `create_folder` takes `parent_folder_id`; without it the folder lands in My Drive root. When the user names a destination folder, resolve its id and pass it - a file created in root when a folder was specified is a wrong result even if the content is perfect. - `create_file` also takes `parent_folder_id`, so create files directly where they belong instead of creating then moving. - The Docs/Sheets/Slides providers' own create tools CANNOT target a folder (their APIs have no parent parameter). When placement matters, either create the artifact there with `create_file` (Markdown to Doc, CSV to Sheet), or create it and then `move_or_rename` with `destination_folder_id`. - `move_or_rename` does either or both in one call; renaming does not change the id. - `update_file` is the same rename/move with the hosted Drive tool's argument names (`fileId`; `title` renames, `parentId` moves by replacing the current parent); it exists on both the translator and hosted surfaces, so prefer it in guidance that must work on either. - Grant limits: this app's grant is `drive.file`, so moving into or writing files/folders this app never created (and the user never picked) can 403. That 403 is a permission fact, not a transient error; do not retry it. ## Markdown and CSV conversion `create_file` with `target_mime_type: application/vnd.google-apps.document` converts Markdown (headings, lists, tables) into a real Google Doc in one call; `...spreadsheet` converts CSV into a Sheet. This is the fastest correct way to produce a formatted Doc: prefer it over create-then-insert when the content is known up front. `update_file_content` converts in place the same way. ## Copying, metadata, and sharing - `copy_file` duplicates a file (optionally into a folder) - the right way to instantiate a template without touching the original. - `get_file_metadata` returns the full record (mime type, parents, size, owners, modified time); use it to confirm a move landed. - `get_file_permissions` lists who has access; `share_file` grants a role to an email. Share only when the user asked; report the link rather than widening access speculatively. - `download_file` returns the raw bytes base64-encoded for non-Google files (images, PDFs, plain text). ## Reading and exporting - `read_file` returns text: Docs as Markdown, Sheets as CSV per tab, Slides as plain text. Use it to verify content after writes. - `export_file` returns rendered bytes base64-encoded: `pdf` for Docs/Sheets/Slides (Sheets PDF renders every tab, charts included), `png`/`svg` for Drawings, plus office formats (docx/xlsx/pptx). Use PDF when the user wants to see the final look; exports cap at 10 MB. - For one slide as PNG use the google-slides `get_slide_thumbnail` tool, not Drive export. ## Comments Comments live on the Drive file, so one tool set covers Docs, Sheets, and Slides alike: - `list_comments` returns each comment's text, quoted document text, resolved state, and replies; it pages, so follow `nextPageToken` before concluding a thread does not exist. - `add_comment` adds a file-level comment (no text anchor). - `reply_to_comment` replies by comment id, and its `action: "resolve"` (content optional) is the ONLY way to resolve a thread; editing document text near a comment never resolves it. Leave already-resolved threads alone. ## Deleting `trash_file` is recoverable (~30 days) and works on files and folders (a trashed folder takes its contents). Permanent deletion is deliberately not available; never promise it. ## Verify your work After creates and moves, confirm placement with `list_folder` on the destination or `get_file_metadata` (check `parents`). After content writes, `read_file` is the ground truth for text and `export_file` (pdf) for the rendered result.