Google Drive (archived)
The gdrive MCP server is Anthropic's archived reference connector for Google Drive, published as @modelcontextprotocol/server-gdrive. It let an assistant search Drive and read file contents, converting Docs to Markdown and Sheets to CSV on the way out. npm now marks the package as no longer supported, and Anthropic moved it to the archived section of the servers repo.
How we started it
npx -y @modelcontextprotocol/server-gdriveSkip it. When we started it with a clean environment, it exited before the initialize handshake, printing that credentials were not found and to run it with an 'auth' argument first, and npm printed a deprecation notice for the package on the way in. That failure is fixable, but you would be spending an afternoon on OAuth client setup for a read-only connector whose maintainer has already walked away. Use Google's own Drive integration or a maintained community server and let this one stay archived.
Use it if
Skip it if
The bill for your context window
We cannot quote a Context Tax for gdrive, because nothing ever got far enough to produce one. A server that dies before initialize never sends a tools list, so our token estimate for it is zero, not small. Every published number on this site comes from a chars/4 estimate over the real tools/list payload, and gdrive never produced one. If you get it authenticated locally, expect a light payload by directory standards: the archived README describes a search tool and file reads, which is a handful of short schemas rather than a sprawling API surface.
Setup reality
Our harness runs every server as an unprivileged user with a clean environment and no secrets, which is exactly the case this one refuses. It scraped no env var names from stderr, because the server does not ask for env vars at all: it wants a Google OAuth client secrets file, a separate run with the 'auth' argument to open a browser consent flow, and the saved token left on disk where the next run can find it. So the real work is in Google Cloud console, enabling the Drive API, creating a desktop OAuth client, downloading the JSON, and choosing a credentials path your agent runner can actually read. None of that is hard. All of it is unpaid maintenance on an archived package.
Questions people ask
Why did the gdrive MCP server fail to start?
It exited with a nonzero code before the initialize handshake, and its last line of stderr said credentials were not found and to run with an 'auth' argument first. It is not a crash. It is a server declining to boot until you have completed a Google OAuth flow and left a token on disk.
Is the gdrive MCP server still maintained?
No. npm prints a deprecation notice saying the package is no longer supported, and Anthropic keeps it in the archived section of the servers repo. Security fixes and Drive API changes are on you once you adopt it.
What should I use instead for Google Drive?
Google's own Drive connector, if your client supports it, or a maintained community Drive server whose repo has commits this year. Either way, measure the tools list before you wire it in, since Drive connectors that add writes and permissions tend to arrive with far bigger schemas.
Other files servers we started
| Server | Vendor | Boots | Context tax |
|---|---|---|---|
| Excel | community (negokaz) | ✓ | ~1,406 tokens |
| Filesystem | Anthropic (reference) | ✓ | ~3,243 tokens |
How this page is made: the server is spawned as an unprivileged user with a clean environment and no credentials, then asked for its tools, resources and prompts over stdio. Token figures are estimates at four characters per token, not a tokenizer count. One run, one machine. Corrections: contact the desk.