A1 AI engineering company GitHub
Connect the Google Drive MCP server to your AI agent Open source, MIT, on npm. Setup instructions for Claude, Cursor and Codex are in the repository; adoption questions — Telegram. GitHub repository Telegram

Google Drive: Download file content — MCP tool

Google Drive MCP tool: Fetches a binary file's bytes — inline for small text, to a local path for everything else.

Technical name: download_file

What task it solves

I want to read or save a file's content.

Downloads the raw bytes of a regular (non-Google-native) Drive file. Small textual files come back inline; anything binary or large is written to a local file.

When to use it

Use it to read a text/CSV/JSON file's content directly in the conversation, or to save a PDF/image/archive to disk. For Google Docs/Sheets/Slides use export_file — they have no bytes to download.

What to provide

  • file_id — required.
  • save_path — optional. Absolute local path; required for binary or >100 KB content.
  • overwrite — optional. Allow replacing an existing local file (default: refuse).
  • acknowledge_abuse — optional. Download a file Drive flagged as malware/abuse (owner only).

What it returns

With save_path: saved_to, bytes, content_type. Inline: content (UTF-8 text) plus name, bytes and content type.

What changes in Google Drive

Nothing in Drive. Locally a file may be written — never over an existing one unless overwrite=true.

Example request

Download the latest export.csv from Google Drive and show me the first rows.

Errors and limitations

Google-native files are rejected with a pointer to export_file; shortcuts are rejected with the real target's id. Inline return is capped at 100 KB and textual content types. Relative save paths are rejected.

Related MCP tools

Technical details