Canvas LMS Integration

Status: Production

Overview

Canvas LMS integration provides real-time academic data for the School domain: courses, assignments, grades, announcements. Data flows through an MCP server to multiple consumers (dashboard, alerts, RAG pipeline).

Data Flow

Canvas LMS API
  │
  ▼
canvas-lms MCP Server (6 tools)
  │
  ├──► canvas_sync.py → assignments JSON + grades YAML + courses YAML
  │
  ├──► grade_alerts.py → Telegram notifications on grade changes
  │
  ├──► deadline-watcher.sh → 48h deadline alerts via Telegram
  │
  ├──► morning-briefing.sh → Academic alerts in daily briefing
  │
  └──► rag_ingest.py → ChromaDB vector store

MCP Server Tools

Tool	Purpose
list_courses	All enrolled courses
list_assignments	Assignments for a course (with due dates, points)
get_assignment	Single assignment details
get_grades	Current grades per course
list_announcements	Recent course announcements
get_course_pages	Course content pages (wiki)

Canvas Sync

Cron: daily at 7 AM
Pulls all assignments across enrolled courses
Saves to local JSON cache
120s timeout via perl alarm wrapper
Handles courses with disabled pages gracefully

Grade Tracking

Per-course: current grade, letter, trend, alerts
Critical threshold: C- or below triggers alert
Grade change detection: compares current vs previous sync
Telegram alert on any grade change

RAG Pipeline

Components

ChromaDB — vector store (persistent local storage)
Ollama nomic-embed-text — embedding model (137M params, local, zero cost)
lightning-whisper-mlx — audio transcription (Apple Silicon optimized)

Scripts

Script	Purpose
rag_ingest.py	512-token recursive chunking, 100 overlap, ingest to ChromaDB
rag_query.py	CLI cross-course semantic search with distance ranking
canvas_to_rag.sh	Cron-ready Canvas pages to ChromaDB sync
transcribe.py	Whisper audio transcription with auto-ingest

Performance

88 chunks ingested across 3 courses
Query accuracy: 0.24 distance for relevant results
5 courses returned 404 on pages (Canvas pages disabled by instructor)