Corpusly vs. Google's Drive connector for Claude

Google's Drive connector is a good fit for pulling up a document you can already name, and if that describes your work, use it. Three differences matter if it doesn't. Formats: the connector's read_file_content tool lists the file types it can turn into text, and plain text, Markdown, CSV, JSON, source code, notebooks, email archives, EPUB, Rich Text, HTML, Google Drawings and SVG are not on that list. Corpusly reads all of them. Search: the connector searches Drive by keyword, so you have to guess the words in the file; Corpusly searches by meaning. Access: the connector exposes file-creation tools, so it asks to write to your Drive, while Corpusly only ever asks to read.

The capability rows are compared against the Google Drive connector's own published tool definitions as of 16 July 2026. “Reads” means extracting a file's text for the assistant to use; the connector can still download other file types as raw data. Google's connector also offers file-creation tools, so it requests write access to your Drive, while Corpusly only ever asks to read. Google runs its connector as a hosted service you cannot inspect; Corpusly's core is source-available, so you can read and run it yourself.

This is the difference most people actually run into. The connector publishes the exact MIME types its read_file_content tool supports, so this table is not an estimate — it is that list, compared against what Corpusly indexes.

A cross means the file type is not named in the connector's own read_file_content supported-types list, so the connector cannot turn it into text for Claude. It does not mean the file is unreachable: the connector's download_file_content tool has no format restriction, so raw file data can still reach Claude. Read from the connector's published tool definitions on 16 July 2026.

The connector's search_files tool uses Google Drive's own keyword grammar — it matches text you supply against the text in your files. There is no semantic or vector retrieval in its published schema. That works when you remember roughly what a document said, and it misses when the document uses different words than your question does.

Corpusly embeds your documents locally and searches them by meaning, so a question about “what we agreed on cancellation” finds the clause that says “termination for convenience”. It then returns the matching passages to Claude, which answers with citations you can open and check.

The connector publishes 8 tools: copy_file, create_file, download_file_content, get_file_metadata, get_file_permissions, list_recent_files, read_file_content, search_files. Two of those — create_file and copy_file — modify your Drive, which is why the connector requests read and write access. Corpusly requests read-only access and cannot modify, move or delete anything in your Drive.

Google's connector is a remote MCP server running at drivemcp.googleapis.com. Your Drive content is extracted on Google's servers and the resulting file text is handed to Claude. Corpusly reads from Drive, parses and embeds on your own machine, and keeps the index there.

Be clear about what that does and doesn't buy you. Corpusly is not a way to keep your documents from your AI provider: the passages it retrieves are sent to Claude and handled under Anthropic's terms. What changes is the scope of exposure— your index, your embeddings, and every file you didn't ask about stay on your device. We make no claim about how Google or Anthropic retain connector data, because we have no first-hand source for it.

If your Drive is mostly Docs, Sheets, Slides, PDFs and Office files; you can usually name or keyword-find the document you want; and you want Claude to create files in Drive as well as read them, the connector does that well, it is already in Claude's directory, and it takes a couple of clicks to enable. Corpusly is for the rest: the formats the connector won't extract, the questions you can't phrase as keywords, and the Drives you'd rather keep on your own machine.

The Drive connector is an MCP server, so its own tools/list response is self-documenting and authoritative — we read the claims off the connector rather than off anyone's marketing. Everything on this page reflects that response as of 16 July 2026. The schema warns that its text representation will change over time, so if you are reading this much later, check it yourself before relying on it, and tell us if we are out of date.

Corpusly is not affiliated with Google or Anthropic. The connector is published by Google, not by Anthropic, and it is a different thing from Anthropic's own built-in Google Workspace connectors.