Documentation

Dropbox connector

Index text files from a Dropbox folder into the knowledge corpus, and use live MCP tools to list, search, and download files. Prefer tenant OAuth (stores a refresh token). A short-lived console access token remains a fallback. Credentials are KMS-encrypted in connector_credentials — never on disk or in env vars.

Connect

  1. In the OrgMCP admin console, open Connectors → Dropbox → Connect. After OAuth, click Test connection. A green result proves the grant can authenticate and list the account root. An empty account stays green — that is a valid empty Dropbox, not a grant problem. A token that cannot list the root fails Test.
  2. Manual fallback: paste a short-lived sl.… access token from the Dropbox app console. It expires; OAuth is the production path.

Ops note: platform DROPBOX_OAUTH_* is live on staging and production (Connect CTA is enabled). Until the Dropbox app leaves Development / Only you, OAuth works only for the app owner. Redirect URIs: https://app.orgmcp.io/api/oauth-tenant/dropbox/callback and https://staging.app.orgmcp.io/api/oauth-tenant/dropbox/callback.

What gets indexed

  • One dropbox_folder source per folder path ('' / / = account root). Recursive by default.
  • Text and source extensions only (.md, .txt, .json, .ts, …). Binaries are skipped.
  • Optional ignorePaths prefixes subtract folders from the walk. Indexing stops at a 1000-file cap and writes a Partial sync warning.

Honesty

  • dropbox_diagnose and admin Test show the authenticated account plus root folder/file counts.
  • A 0-chunk crawl writes an exact emptyReason (empty folder, path filter, only binaries, oversized files) — not a generic “check grants” line.
  • The folder lister rethrows API errors; it never swallows them into an empty list.

Sources UI

The Index tab uses the flat snapshot list (same surface as Notion / Jira), not a dedicated folder-tree picker. Pick a folder, optionally turn off recursion or add ignore paths, then sync.