From your university account
to the AI tools you use.
UnivMCP separates university authentication and collection from AI retrieval. A local collector observes authorized LMS resources and pushes normalized changes to the cloud. Connected agents retrieve course context from that synchronized state.
This page describes the system we are building. The first target is Chung-Ang University e-Class on macOS. A live Chung-Ang exam timetable search already runs at univmcp.com/cau/exam.
- 01 / SOURCEUniversity LMSYour courses and their original resources.
- 02 / ON YOUR DEVICELocal CollectorUniversity authentication, resource collection, and outbound sync.
- 03 / SYNCHRONIZEDUnivMCP CloudCourse metadata, original files, and access for your paired AI connection.
- 04 / WHEN NEEDEDMCP / API → AIConnected tools retrieve your synchronized course context.
Connect locally. Retrieve remotely.
Local credential custody
University sign-in belongs to the local collector. Persistent credentials use Keychain-backed storage in the macOS design. Device credentials authenticate cloud synchronization separately from the university account.
Context-preserving synchronization
Original files are stored with normalized course and resource metadata. LMS-native identifiers preserve the relationship between each file, its course, and its observed version.
Outbound-only cloud sync
The collector pushes changes over outbound HTTPS connections. Agents read from the cloud, so resource retrieval requires no inbound port or tunnel to the laptop.
Fingerprint-based delta sync
Resource fingerprints identify changes in upstream metadata and determine when acquisition needs to run again. SHA-256 hashes identify downloaded bytes. The sync engine uploads missing or changed files and upserts metadata; idempotent writes allow retries without duplicating logical state.
Principal-scoped access
Each AI connection resolves to one principal. Resource access is established by that principal’s authenticated collector observations and checked on retrieval. Storage reuse does not grant access to another user’s resources.
Scheduled collection
Collection runs on app startup, device wake, periodic intervals while the device is active, and manual refresh. Retrieval uses the last synchronized state. Sync status makes freshness inspectable.
Remote MCP over synchronized state.
The planned remote Model Context Protocol (MCP) interface exposes four initial operations: list courses, list course resources, retrieve a resource, and inspect sync status. The retrieval API reads the same normalized cloud state.
University authentication and LMS-specific response handling stay in the local adapter. Additional compatible clients can use the shared retrieval interface without implementing a university integration.
Chung-Ang e-Class → macOS → AI.
The first milestone is a daily-usable flow from local collection through cloud synchronization to remote resource retrieval. Course resources come first; assignments, announcements, additional university adapters, and other desktop platforms can follow as the core path matures.
Where the project began
I use cau-eclass-mcp with Codex for my own courses at Chung-Ang University. That work provides verified LMS behavior, resource normalization rules, and regression fixtures. I’m carrying those lessons into UnivMCP’s dedicated local collector, cloud sync engine, and remote retrieval interface.