Summary
An agent can work a big mind map without swallowing it.
Instead of returning the whole JSON tree, it hands back an outline with entry points and lets the model drill in by node path — and when a chosen section is still too big, it paginates rather than truncating silently.
What it is
A reader for MindMup 2 .mup files stored in Google Drive. Small maps come back whole; large ones return a tree outline with suggested entry points, and you drill in by node path — so a 3MB map never lands in the model's context in one piece.
What you get
- MindMup files listed from Drive with id, name, folder URL, size and modified time
- A map read by file id or by partial name — full content when it is small, an outline with section statistics and suggested starting paths when it is large
- Keyword search across nodes, scopable to a subtree, returning node path, title preview and breadcrumb
- A section fetched by dotted node path with depth, offset and limit, paginating or truncating when the section is itself too large
- Per-client cache isolation keyed on a client id header, so two users never share cached content
Requirements
A Google Cloud service account with Viewer access to the Drive folder, its JSON key base64-encoded and passed as a header, and a client that speaks HTTP transport — Claude Code or Cursor. Claude Desktop is stdio-only and is not supported.