google-drive-files-mcp
A focused MCP server for Google Drive file management, Google Sheets editing, and surgical Google Docs editing with 23 granular, single-purpose tools.
README
google-drive-files-mcp
<!-- mcp-name: io.github.zayansalman/google-drive-files-mcp -->
A focused Model Context Protocol server (and standalone CLI) for Google Drive file management (move, upload, organize), Google Sheets editing (cells, rows, tabs, format), and surgical Google Docs editing (replace a phrase, insert at an anchor, delete a range — leaving all other formatting untouched). Twenty-three granular, single-purpose tools — each an explicit operation, with writes returning before/after or precise indices so nothing changes silently.
Built because the hosted Google Drive connectors can search, read, and copy files, but cannot move them (copy duplicates a file with a new ID rather than relocating it), cannot reliably upload a real binary (their inline-content API needs the bytes embedded as base64, which breaks for large/binary files), and cannot edit spreadsheet or document contents in place. This server adds a true move, a resumable upload from a local path, folder creation, a full set of Google Sheets edits, and range-scoped Google Docs edits — for any MCP client (Claude Code, Claude Desktop, Cursor, Cline, etc.).
Tools
Drive — files & folders
drive_search(query, only_folders=False, max_results=20)— find files/folders (and their IDs).drive_create_folder(name, parent=None)— create a folder (parentaccepts an ID, URL,root, or an unambiguous folder name).drive_move(file, dest_folder, keep_existing_parents=False)— true move (removed from its current folder);keep_existing_parents=Trueadds without removing.drive_upload_file(local_path, parent=None, name=None, mime_type=None)— upload a local file (any type, incl. large binaries) via a resumable media upload. Stored as-is (no Google-format conversion).
Google Sheets — read, edit, structure, format
sheets_get_info(spreadsheet)— list tabs + sizes (call first to learn tab names).sheets_read(spreadsheet, range_a1, render_option="FORMATTED_VALUE")— read a range (FORMULA/UNFORMATTED_VALUEavailable).sheets_write(spreadsheet, range_a1, values, value_input_option="USER_ENTERED")— overwrite a range; returnsbefore.USER_ENTEREDparses numbers and makes=SUM(..)a formula;RAWwrites literally.sheets_append(spreadsheet, range_a1, values, ...)— append rows after a table.sheets_clear(spreadsheet, range_a1)— clear values (keeps formatting); returnsbefore.sheets_batch_write(spreadsheet, updates, ...)— write many ranges atomically.sheets_add_tab/sheets_rename_tab/sheets_delete_tab— manage tabs (delete is destructive).sheets_freeze_panes(spreadsheet, tab, rows=None, cols=None)— freeze/unfreeze header rows/cols (0 = explicitly unfreeze; omit = untouched).sheets_format(spreadsheet, range_a1, number_format=None, bold=None, background=None, wrap=None)— number pattern / bold /#RRGGBBbackground / text wrap.sheets_set_column_width(spreadsheet, range_a1, pixel_size)— set column width in pixels for a whole-column range (e.g.'A:C').sheets_set_row_height(spreadsheet, range_a1, pixel_size)— set row height in pixels for a whole-row range (e.g.'2:5').
Google Docs — surgical in-place editing
Every Docs edit is range-scoped (documents.batchUpdate), so it changes exactly what you target and leaves every other run's native formatting byte-for-byte intact.
docs_get_info(document)— title + structure (paragraph/table counts, end index).docs_read(document)— full text with per-segment character indices (table cells include row/col). Read this first to target edits.docs_replace_text(document, find, replace, match_case=False)— replace a phrase everywhere; surrounding content/formatting untouched.docs_insert_text(document, text, index=None, after_text=None, before_text=None)— insert at an index or right after/before an anchor phrase.docs_delete_range(document, start_index, end_index)— delete one precise range.docs_append_text(document, text)— append at the end.
No rename/copy/trash/delete of files (copy already exists in the hosted Drive connector); destructive ops (sheets_clear, sheets_delete_tab, docs_delete_range) are their own explicit tools so you control exactly when they run.
Scope warning (read this)
Moving an existing arbitrary file requires the full drive scope — read/write/delete on all your Drive files. There is no narrower scope that can change a file's parents (drive.file only covers files the app itself created). This tool is therefore far more powerful than a read-only connector. Treat its cached token like a password (stored 0600), and prefer an isolated OAuth client/token rather than sharing one with read-only tools.
Install
pip install google-drive-files-mcp
# or: uv tool install google-drive-files-mcp
One-time setup (~10 min)
- Google Cloud Console → create/pick a project.
- Enable the Drive API (link); for the Sheets tools also enable the Sheets API (link); for the Docs tools also enable the Docs API (link). (The full
drivescope already authorizes the Sheets and Docs APIs — no extra consent, just enable each API you want.) - OAuth consent screen → Internal (Workspace; no verification needed even for the restricted full-drive scope) or External + add yourself as a test user (personal Gmail).
- Credentials → OAuth client ID → Desktop app → download JSON.
google-drive-files-mcp setup --import-credentials ~/Downloads/client_secret_*.json→ consent in the browser.
Verify:
google-drive-files-mcp status
google-drive-files-mcp search "Reports" --folders
Claude Code
claude mcp add --scope user google-drive-files google-drive-files-mcp -- serve
Find my "Q2 Report" doc and move it into the "2026 Reports" folder.
Claude Desktop
{
"mcpServers": {
"google-drive-files": { "command": "google-drive-files-mcp", "args": ["serve"] }
}
}
CLI
google-drive-files-mcp search "budget" # find a file + its ID
google-drive-files-mcp search "Reports" --folders # find destination folders
google-drive-files-mcp mkdir "2026 Reports" # create a folder
google-drive-files-mcp move <file-url-or-id> "2026 Reports" # move into it (by folder name)
google-drive-files-mcp move <file-url-or-id> root # move back to My Drive root
google-drive-files-mcp upload ~/Downloads/report.zip --parent "2026 Reports" # upload a local file
# Google Sheets
google-drive-files-mcp sheet-info <sheet-url-or-id> # list tabs + sizes
google-drive-files-mcp sheet-read <sheet-url-or-id> "Sheet1!A1:D10" # read a range
google-drive-files-mcp sheet-write <sheet-url-or-id> "Sheet1!B2:B4" '[[100],[200],[300]]' # set numbers
google-drive-files-mcp sheet-format <sheet-url-or-id> "Sheet1!B2:B4" --number-format '#,##0.00' --bold
google-drive-files-mcp sheet-format <sheet-url-or-id> "Sheet1!A1:B2" --wrap # wrap text
google-drive-files-mcp sheet-freeze-panes <sheet-url-or-id> Sheet1 --rows 1 # freeze header row
google-drive-files-mcp sheet-set-column-width <sheet-url-or-id> "Sheet1!A:C" 150 # column width (px)
google-drive-files-mcp sheet-set-row-height <sheet-url-or-id> "Sheet1!2:5" 40 # row height (px)
# Google Docs (surgical, in-place)
google-drive-files-mcp doc-read <doc-url-or-id> # text + character indices
google-drive-files-mcp doc-replace <doc-url-or-id> "old phrase" "new phrase" # preserves all other formatting
google-drive-files-mcp doc-insert <doc-url-or-id> $'\n\nNew paragraph.' --after "some anchor text"
drive_move returns the before/after parents so the move is auditable:
{ "id": "1Abc…", "name": "Q2 Report", "moved_from": ["0OldFolderId"], "moved_to": "1NewFolderId", "parents_now": ["1NewFolderId"], "web_view_link": "https://docs.google.com/…" }
Configuration
| Variable | Default | What |
|---|---|---|
GDRIVE_FILES_MCP_CREDENTIALS |
~/.config/google-drive-files-mcp/credentials.json |
OAuth client secret |
GDRIVE_FILES_MCP_TOKEN |
~/.config/google-drive-files-mcp/token.json |
Cached refresh token |
GDRIVE_FILES_MCP_SCOPES |
https://www.googleapis.com/auth/drive |
OAuth scopes (comma-separated) |
Safety notes
- True move removes the file from its current folder. Drive items can have multiple parents;
keep_existing_parents=Trueadds without removing. - Folder-name destinations must be unambiguous. If two folders share a name,
drive_moverefuses and lists the candidates so you can pass an explicit ID. - The before/after parents are returned on every move so an agent (or you) can verify the result.
Troubleshooting
HttpError 403: Google Drive API has not been used in project … or it is disabled
Enable the Drive API on the project that owns your OAuth client: console.cloud.google.com/apis/library/drive.googleapis.com → Enable, then wait ~1 min and retry. (This is the single most common first-run error.)
multiple folders named '…'; pass the folder ID/URL instead
Two or more of your folders share that name, so a name is ambiguous. Run google-drive-files-mcp search "<name>" --folders to get the IDs and pass the exact one.
no folder named '…' found
Create it first (google-drive-files-mcp mkdir "<name>") or pass a folder ID/URL/root.
HttpError 403: insufficientFilePermissions when moving
Either the token lacks the full drive scope — re-authorize with google-drive-files-mcp setup --reauth — or you don't have edit rights on the file/destination (you can't move a file someone else owns unless they've granted you edit access).
No valid Google token from Claude Desktop / cron
The first consent needs a browser. Run google-drive-files-mcp setup once in a terminal; later headless runs reuse and auto-refresh the cached token.
A move "removed" a file from a shared folder unexpectedly
A true move removes the item from its current parent(s). If you meant to also keep it where it was, use keep_existing_parents=true (CLI --keep).
Use it from other clients (Cursor, Cline, Continue, …)
Any stdio MCP client works — see docs/other-clients.md.
Authentication — bring your own Google OAuth client
There are no API keys and no shipped secrets. The server authenticates to your Google account with an OAuth client you create, and caches a refresh token locally. The author has zero access to your data.
- Why your own client? Google's restricted scopes (here, the full
drivescope) can't be redistributed in a shared app, and an unverified shared app is capped at 100 users. "Bring your own OAuth client" is the standard pattern for personal-data MCP servers. - What you need: a free Google Cloud project, the Drive API enabled, an OAuth consent screen, and a Desktop OAuth client. Full walkthrough → docs/setup-google-oauth.md.
- Where your token lives:
~/.config/google-drive-files-mcp/token.json(mode0600). Delete it to revoke locally; revoke fully at myaccount.google.com/permissions. - No hosted/SaaS option — everything runs locally; your Drive data never touches a third-party server.
More guides
- docs/setup-google-oauth.md — full OAuth walkthrough (full-
drivescope) + common errors - docs/claude-code.md · docs/claude-desktop.md · docs/other-clients.md
- examples/ — e.g. file a document into a
YYYY-MMfolder
Related tools
Part of a small family of focused, local MCP servers for Google Workspace data the hosted connectors don't expose:
- gmail-attachments-mcp — download Gmail attachment bytes to disk
- google-drive-comments-mcp — read comment threads on Docs/Sheets/Slides
- google-drive-files-mcp — move/organize Drive files (this repo)
They can share one OAuth login or stay isolated — see each repo's setup.
License
MIT. See LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.