canvas-bulk-grading

Canvas Bulk Grading

Grade Canvas LMS assignments efficiently using rubric-based workflows. This skill requires the Canvas MCP server to be running and authenticated with an instructor or TA token.

Prerequisites

• Canvas MCP server running and connected
• Authenticated with an educator (instructor/TA) Canvas API token
• Assignment must exist and have submissions to grade
• Rubric must already be created in Canvas and associated with the assignment (Canvas API cannot reliably create rubrics -- use the Canvas web UI for that)

Workflow

Step 1: Gather Assignment and Rubric Information

Before grading, retrieve the assignment details and its rubric criteria.

get_assignment_details(course_identifier, assignment_id)

Then get the rubric. Use get_assignment_rubric_details if the rubric is already linked to the assignment, or list_all_rubrics to browse all rubrics in the course:

get_assignment_rubric_details(course_identifier, assignment_id)
list_all_rubrics(course_identifier)
get_rubric_details(course_identifier, rubric_id)

Record the criterion IDs (often prefixed with underscore, e.g., _8027) and rating IDs from the rubric response. These are required for rubric-based grading.

Step 2: List Submissions

Retrieve all student submissions to determine how many need grading:

list_submissions(course_identifier, assignment_id)

Note the user_id for each submission and the workflow_state (submitted, graded, pending_review). Count the submissions that need grading to determine which strategy to use.

Step 3: Choose a Grading Strategy

Use this decision tree based on the number of submissions to grade:

How many submissions need grading?
|
+-- 1-9 submissions
|   Use grade_with_rubric (one call per submission)
|
+-- 10-29 submissions
|   Use bulk_grade_submissions (concurrent batch processing)
|   Set max_concurrent: 5, rate_limit_delay: 1.0
|   ALWAYS run with dry_run: true first
|
+-- 30+ submissions OR custom grading logic needed
    Use execute_typescript with bulkGrade function
    Grading logic runs locally; only selected output returns to the model
    ALWAYS run with dry_run: true first