Gdrive Cart

GDrive Context Cart

A lightweight visual context staging bridge connecting cloud storage and AI coding agents (Hermes, Claude Code, Cursor, OpenClaw, Codex, etc.).

---

English

Overview

Most agent integrations for cloud drives (such as headless CLI tools or MCP servers) rely exclusively on programmatic API calls or keyword searches. When a workspace contains thousands of documents, an agent searching blind wastes significant context tokens and frequently retrieves outdated or incorrect revisions.

`gdrive-cart` implements a **Context Staging Basket** architecture:

1. Human Curation: Rapid visual navigation, search, and selection of relevant files and folders in a clean local web interface.
2. Machine Processing: Queued batch downloading, automatic format conversion (Docs to Markdown, Sheets to CSV), recursive directory resolution, and compilation of a structured _context_manifest.md.
3. Seamless Hand-off: Automatic generation and copying of the local path hand-off text to the system clipboard, allowing users to paste (Ctrl + V) directly into their agent interface without disrupting prompt composition.

Upstream Dependency & Companion Architecture

`gdrive-cart` is designed as a **companion visual tool** rather than a standalone authentication stack. It builds directly upon the official **`google-workspace`** skill (by Nous Research):

• Authentication Delegation: It intentionally does not re-implement the Google Cloud OAuth PKCE verification flow. Instead, it delegates credential creation and refresh to the upstream google-workspace setup, consuming the resulting google_token.json.
• Separation of Concerns: The upstream google-workspace skill provides headless function-calling tools; gdrive-cart provides the human-facing visual staging layer, format conversion, and context distillation.
• Standalone Usage: In non-Hermes environments (such as Claude Code or Cursor), users simply place an authorized `google_token.j