canvas-peer-review-manager（画布同行评审经理）

Canvas Peer Review Manager

A complete peer review management workflow for educators using Canvas LMS. Monitor completion, analyze quality, identify students who need follow-up, send reminders, and export data -- all through MCP tool calls against the Canvas API.

Prerequisites

• Canvas MCP server must be running and connected to the agent's MCP client (e.g., Claude Code, Cursor, Codex, OpenCode).
• The authenticated user must have an educator or instructor role in the target Canvas course.
• The assignment must have peer reviews enabled in Canvas (either manual or automatic assignment).
• FERPA compliance: Set ENABLE_DATA_ANONYMIZATION=true in the Canvas MCP server environment to anonymize student names. When enabled, names render as Student_xxxxxxxx hashes while preserving functional user IDs for messaging.

Steps

1. Identify the Assignment

Ask the user which course and assignment to manage peer reviews for. Accept a course code, Canvas ID, or course name, plus an assignment name or ID.

If the user does not specify, prompt:

Which course and assignment would you like to check peer reviews for?

Use list_courses and list_assignments to help the user find the right identifiers.

2. Check Peer Review Completion

Call get_peer_review_completion_analytics with the course identifier and assignment ID. This returns:

• Overall completion rate (percentage)
• Number of students with all reviews complete, partial, and none complete
• Per-student breakdown showing completed vs. assigned reviews

Key data points to surface:

3. Review the Assignment Mapping

If the user wants to understand who is reviewing whom, call get_peer_review_assignments with:

• include_names=true for human-readable output
• include_submission_details=true for submission context

This shows the full reviewer-to-reviewee mapping with completion status.

4. Extract and Read Comments

Call get_peer_review_comments to retrieve actual comment text. Parameters:

• include_reviewer_info=true -- who wrote the comment
• include_reviewee_info=true -- who received the comment
• anonymize_students=true -- recommended when sharing results or working with sensitive data

This reveals what students actually wrote in their reviews.

5. Analyze Comment Quality

Call analyze_peer_review_quality to generate quality metrics across all reviews. The analysis includes:

• Average quality score (1-5 scale)
• Word count statistics (mean, median, range)
• Constructiveness analysis (constructive feedback vs. generic comments vs. specific suggestions)
• Sentiment distribution (positive, neutral, negative)
• Flagged reviews that fall below quality thresholds

Optionally pass analysis_criteria as a JSON string to customize what counts as high/low quality.

6. Flag Problematic Reviews

Call identify_problematic_peer_reviews to automatically flag reviews needing instructor attention. Flagging criteria include:

• Very short or empty comments
• Generic responses (e.g., "looks good", "nice work")
• Lack of constructive feedback
• Potential copy-paste or identical reviews

Pass custom criteria as a JSON string to override default thresholds.

7. Get the Follow-up List

Call get_peer_review_followup_list to get a prioritized list of students requiring action:

• priority_filter="urgent" -- students with zero reviews completed
• priority_filter="medium" -- students with partial completion
• priority_filter="all" -- everyone who needs follow-up
• days_threshold=3 -- adjusts urgency calculation based on days since assignment

8. Send Reminders

Always use a dry run or review step before sending messages.

For targeted reminders, call send_peer_review_reminders with:

• recipient_ids -- list of Canvas user IDs from the analytics results
• custom_message -- optional custom text (a default template is used if omitted)
• subject_prefix -- defaults to "Peer Review Reminder"

Example flow:

1. Get incomplete reviewers from step 2
2. Extract their user IDs
3. Review the recipient list with the user
4. Send reminders after confirmation

For a fully automated pipeline, call send_peer_review_followup_campaign with just the course identifier and assignment ID. This tool:

1. Runs completion analytics automatically
2. Segments students into "urgent" (none complete) and "partial" groups
3. Sends appropriately toned reminders to each group
4. Returns combined analytics and messaging results

Warning: The campaign tool sends real messages. Always confirm with the instructor before running it.

9. Export Data

Call extract_peer_review_dataset to export all peer review data for external analysis:

• output_format="csv" or output_format="json"
• include_analytics=true -- appends quality metrics to the export
• anonymize_data=true -- recommended for sharing or archival
• save_locally=true -- saves to a local file; set to false to return data inline

10. Generate Instructor Reports

Call generate_peer_review_feedback_report for a formatted, shareable report:

• report_type="comprehensive" -- full analysis with samples of low-quality reviews
• report_type="summary" -- executive overview only
• report_type="individual" -- per-student breakdown
• include_student_names=false -- recommended for FERPA compliance

For a completion-focused report (rather than quality-focused), use generate_peer_review_report with options for executive summary, student details, action items, and timeline analysis. This report can be saved to a file with save_to_file=true.

Use Cases

"How are peer reviews going?" Run steps 1-2. Present completion rate, highlight any concerning patterns (e.g., "Only 60% complete, 8 students haven't started").

"Who hasn't done their reviews?" Run steps 1-2, then step 7 with priority_filter="urgent". List the students who need follow-up.

"Are the reviews any good?" Run steps 4-6. Present quality scores, flag generic or low-effort reviews, and surface recommendations.

"Send reminders to stragglers" Run steps 1-2 to identify incomplete reviewers, then step 8. Always confirm the recipient list before sending.

"Give me a full report" Run steps 2, 5, 6, and 10. Combine completion analytics with quality analysis into a comprehensive instructor report.

"Export everything for my records" Run step 9 with output_format="csv" and anonymize_data=true for a FERPA-safe dataset.

MCP Tools Used

Example

User: "How are peer reviews going for Assignment 3 in BADM 350?"

Agent: Calls get_peer_review_completion_analytics and presents:

## Peer Review Status: Assignment 3

- **Completion rate:** 72% (23/32 students fully complete)
- **Partial:** 5 students (started but not finished)
- **Not started:** 4 students

### Students Needing Follow-up
**Not started (urgent):**
- Student_a8f7e23 (0 of 3 reviews done)
- Student_b2c91d4 (0 of 3 reviews done)
- Student_f5e67a1 (0 of 3 reviews done)
- Student_d9c34b2 (0 of 3 reviews done)

**Partial (needs nudge):**
- Student_c1d82e5 (1 of 3 reviews done)
- Student_e4f03a9 (2 of 3 reviews done)

User: "Send reminders to the ones who haven't started"

Agent: Confirms the 4 recipients, then calls send_peer_review_reminders with their user IDs.

User: "Now check if the completed reviews are any good"

Agent: Calls analyze_peer_review_quality and presents quality scores, flags 3 reviews as too short, and recommends the instructor follow up with specific students.

Safety Guidelines

• Confirm before sending -- Always present the recipient list and message content to the instructor before calling any messaging tool.
• Use dry runs -- When testing workflows, start with a single recipient or confirm the output of analytics tools before acting on the data.
• Anonymize by default -- Use anonymize_students=true or anonymize_data=true when reviewing data in shared contexts.
• Respect rate limits -- The Canvas API allows roughly 700 requests per 10 minutes. For large courses, the messaging tools send messages sequentially with built-in delays.
• FERPA compliance -- Never display student names in logs, shared screens, or exported files unless the instructor has explicitly confirmed the context is appropriate.

Notes

• Peer reviews must be enabled on the assignment in Canvas before any of these tools return data.
• The send_peer_review_followup_campaign tool combines analytics and messaging into one call -- powerful but sends real messages. Use it only after confirming intent with the instructor.
• Quality analysis uses heuristics (word count, keyword matching, sentiment). It identifies likely low-quality reviews but is not a substitute for instructor judgment.
• This skill pairs well with canvas-morning-check for a full course health overview that includes peer review status alongside submission rates and grade distribution.