Skip to main content
Glama

canvas-mcp-lite

A lean, instructor-focused MCP server for Canvas LMS, built with FastMCP. It lets an AI assistant act as a teaching assistant against your Canvas instance: browsing courses, reading student submissions (including PDF/DOCX file uploads), grading with rubrics, posting announcements, managing modules and pages, and more.

Tools

94 tools across 15 domains, organized by risk level:

Group

Count

Examples

Read

45

list_courses, list_ungraded_submissions, get_submission_content, get_submission_forensics, get_submission_annotations, read_google_slides, get_new_quiz_item_analysis

Write

39

create_assignment, grade_submission, grade_with_rubric, comment_on_google_doc, edit_google_doc_text, update_google_slide_text, create_google_slides, randomly_assign_peer_reviews, upload_course_file

Delete

10

delete_assignment, delete_page, bulk_delete_announcements, delete_google_slide

Highlights:

  • Read what students actually submitted — get_submission_content extracts text from uploaded PDFs, DOCX files, and plain text (plus typed text entries, URLs, and discussion submissions) and includes the submission comment thread, so grading tools work from real content.

  • A real grading workflow — list_ungraded_submissions is the grading queue, grade_submission handles points, pass/fail, letter, and percent grades (and reports partial success when Canvas saves a comment but rejects a grade), and post_grades/hide_grades control when students see results.

  • Course codes or IDs — every course-scoped tool accepts either a numeric course ID or a course_code string (resolved with a short-TTL cache).

  • Safe by default — assignments and pages are created unpublished unless you say otherwise; destructive tools are clearly marked.

  • Peer review, end to end — assign reviews randomly (randomly_assign_peer_reviews, a fair shuffled rotation) or from your own pairings/groups (assign_peer_reviews_manual); both only involve students who submitted their own draft — no one reviews (or is reviewed) if they didn't turn in their draft, so no one earns reviewer credit without doing their own work. Then get_submission_annotations and summarize_reviewer_annotations read the highlights and margin comments reviewers leave in Canvas's DocViewer (which never appear in the normal comment thread), so you can grade the quality of each student's peer feedback. All with dry-run previews.

  • LLM-friendly output — every tool returns formatted, readable text rather than raw JSON.

  • Grade Google Docs in place — when students submit Google Doc links, read_google_doc reads the live document and comment_on_google_doc leaves feedback as real Google Docs comments from the instructor's account (see below).

  • Read, edit, and build Google Slides decks — the same connected account can find the instructor's existing lecture decks, read them slide by slide, edit text in place, add/move/duplicate/delete slides, set speaker notes, copy a deck as a template, create a new deck, and comment on student Slides submissions (see below).

Related MCP server: Canvas MCP Server

Google Docs and Google Slides (optional)

Students in writing courses often submit a Google Doc link rather than a file — commonly pasted as a submission comment. The same connection also lets the instructor work on their own Google Slides lecture decks from chat (find, read, edit in place, build new ones) — see the Google Slides note below. With a Google account connected, the grading loop becomes: list_google_doc_links (one call collects every student's doc link from their submission comments or URL submissions, and flags who hasn't posted one) → read_google_doc (read the live draft) → comment_on_google_doc (one call per piece of feedback, quoting the passage it refers to) → grade_submission (score in Canvas).

Connecting a Google account

Remote/HTTP servers — in-chat, no terminal (recommended). The instructor asks Claude to run google_docs_status / connect_google_docs, opens the sign-in link it returns in their browser, and approves. The redirect lands on the server's public /oauth/google/callback route, which activates commenting immediately and shows a GOOGLE_OAUTH_REFRESH_TOKEN value to save in the server's environment (Railway → Variables) so the connection survives restarts. Completing the flow requires a single-use state minted by the tool (which sits behind the secret MCP_PATH), so the public callback can't be abused to swap accounts.

One-time admin prerequisite at console.cloud.google.com: create a project, enable the Google Drive API, Google Docs API, and Google Slides API, configure and publish the OAuth consent screen (unpublished "Testing" apps expire refresh tokens after 7 days), create an OAuth client ID of type Web application with authorized redirect URI https://<server-domain>/oauth/google/callback, and set GOOGLE_OAUTH_CLIENT_ID and GOOGLE_OAUTH_CLIENT_SECRET in the server's environment.

Local stdio servers — terminal. Create a Desktop app OAuth client instead, run canvas-mcp-google-auth, and sign in when the browser opens (use --manual if the terminal is on a different machine than the browser). It saves all three GOOGLE_OAUTH_* values to .env.

Notes:

  • Comments post as the connected Google account — connect the instructor's account, not a bot.

  • The instructor's account needs at least Commenter access to each student doc (have students share their docs with the instructor, or use "Anyone with the link — Commenter").

  • comment_on_google_doc takes a quoted_text passage and pins the comment to that text in the document, so the student sees the passage highlighted with the note in the margin, exactly like a comment left by hand. Repeated passages take an occurrence argument. Anchoring uses the Google Docs API's insertComment, which is currently in Developer Preview — if the connected account isn't enrolled, the comment still posts, just unanchored with the passage quoted in the card, and the tool's reply says which happened. Enrolling needs no new consent: the existing auth/drive scope covers it. The Google Docs API must be enabled in the same Cloud project as the Drive API (APIs & Services → Library), or every quoted comment falls back to unanchored with an HTTP 403 note.

  • list_new_quizzes / get_new_quiz_item_analysis cover New Quizzes (the classic list_quizzes tools don't). Item analysis asks Canvas to build its item-analysis report, waits for it, and returns the most-missed questions ranked plus per-question correct/incorrect counts, mean, difficulty and discrimination indices, and answer-choice distributions (per prompt for matching items).

  • Grading tools (grade_submission, grade_with_rubric) attach comments to the student's most recent attempt and warn in the reply when a student has submitted more than once; get_submission_content and list_ungraded_submissions flag resubmissions too.

  • Editing Google Docs directly. edit_google_doc_text replaces one quoted passage with new text (loose matching on whitespace, smart quotes, and case; occurrence picks a repeat; empty replacement deletes), insert_text_in_google_doc appends a new paragraph at the end or inserts before/after a quoted anchor, and replace_text_in_google_doc is a document-wide find-and-replace. These are direct edits made as the connected account — the Docs API has no suggesting mode, so they are not tracked changes — and need Editor access. Feedback on student writing should still go through comment_on_google_doc; the edit tools are for the instructor's own documents (syllabi, handouts, templates) or when an edit to a student doc is explicitly wanted.

  • get_google_doc_forensics reports Draftback-style revision-history facts for a shared doc — editing sessions, time span, text typed in small edits vs. added in large single insertions (often pastes), and who edited. Facts only, no AI-detection verdicts; it degrades to a coarse timeline if the detailed changelog is unavailable.

  • Google Slides. list_google_slides finds decks the connected account can see (optionally by title), and read_google_slides prints every slide with its title, text boxes, bullet lists, table cells, images, and speaker notes — each tagged with the object ID the edit tools take. Edits happen in the existing deck, keeping its theme and formatting: update_google_slide_text (one text box or table:row:col cell; lines starting with - become bullets), replace_text_in_google_slides (deck-wide find-and-replace, e.g. dates or {{placeholders}}), update_google_slide_notes, add_google_slide (layouts: title_and_body, title_only, title, section_header, blank), move_google_slide, duplicate_google_slide, and delete_google_slide. copy_google_slides clones a deck as a starting point and create_google_slides builds a new one from a JSON list of slides. Student decks shared with the account work the same way for reading, plus list_google_slides_comments / comment_on_google_slides (Drive comments — the Slides API can't pin a comment to a text range, so name the slide in the comment). Editing needs Editor access on the deck; the Google Slides API must be enabled in the Cloud project.

  • The scope requested is full Drive access (auth/drive) — the narrower scopes can't comment on files the app didn't create. The token lives only in the server's environment.

Setup

Requires Python 3.10+.

  1. Install (editable, from the project root):

    python -m venv .venv
    source .venv/bin/activate
    pip install -e .
  2. Configure credentials in a .env file next to the package:

    CANVAS_API_URL=https://yourschool.instructure.com/api/v1
    CANVAS_API_TOKEN=your-canvas-access-token

    Generate a token in Canvas under Account → Settings → New Access Token. The .env is loaded relative to the package location (not the working directory), because MCP clients launch servers from arbitrary directories.

  3. Run via the console script:

    canvas-mcp-lite

Using with Claude

Register it as an stdio MCP server. For Claude Code:

claude mcp add canvas -- /path/to/.venv/bin/canvas-mcp-lite

Or in a Claude Desktop / MCP client config:

{
  "mcpServers": {
    "canvas": {
      "command": "/path/to/.venv/bin/canvas-mcp-lite"
    }
  }
}

Project layout

canvas_mcp_lite/
  server.py      FastMCP entry point; registers READ/WRITE/DELETE tool lists
  client.py      Async Canvas API client: auth, pagination, retry, errors
  google_client.py  Google Drive/Docs/Slides API client (refresh-token auth, file-ID parsing)
  google_auth.py    One-time OAuth setup CLI (canvas-mcp-google-auth)
  util.py        Course code → ID resolution (cached), date formatting
  tools/         One module per domain: courses, assignments, grading,
                 modules_pages, announcements, discussions, files, quizzes,
                 messaging, peer_review, analytics, integrity, google_docs,
                 google_slides, new_quizzes

Notes

  • The package name is canvas_mcp_lite and it uses relative imports — run it through the installed canvas-mcp-lite script, not python server.py.

  • File downloads for text extraction are capped at 25MB; scanned/image-only PDFs return a notice instead of text.

  • Canvas API errors surface with status code and URL; transient timeouts are retried with backoff.

Available Tools

64 tools
add_module_itemA

Add an item to a module. item_type: File, Page, Discussion, Assignment, Quiz, SubHeader, ExternalUrl, ExternalTool. Page items need page_url; most others need content_id; ExternalUrl needs external_url.

Inputs: item_type (required), module_id (required), course_identifier (required), title, +5 more. Returns: result

assign_peer_reviewA

Assign reviewer_user_id to peer-review reviewee_user_id's submission on this assignment. The assignment must have peer_reviews=True set first (see update_assignment).

Inputs: assignment_id (required), reviewee_user_id (required), reviewer_user_id (required), course_identifier (required). Returns: result

bulk_delete_announcementsA

Delete multiple announcements at once. announcement_ids is a comma-separated list of IDs (get them from list_announcements first). This cannot be undone.

Inputs: announcement_ids (required), course_identifier (required). Returns: result

bulk_grade_submissionsA

Post grades for many students at once on one assignment. grades: {user_id: score, ...}. Same optional comment applied to every submission. This is a single high-blast-radius call — double check the grades dict before calling.

Inputs: grades (required), assignment_id (required), course_identifier (required), comment. Returns: result

create_announcementA

Create an announcement. Pass delayed_post_at (ISO-8601) to schedule it instead of posting now.

Inputs: title (required), message (required), course_identifier (required), delayed_post_at. Returns: result

create_assignmentB

Create an assignment. due_at is ISO-8601 (e.g. 2026-10-09T23:59:59-04:00). Defaults to unpublished. submission_types is a comma-separated string, e.g. "online_upload,online_text_entry" or "on_paper".

Inputs: name (required), course_identifier (required), due_at, published, +3 more. Returns: result

create_discussion_topicC

Create a new (non-announcement) discussion topic.

Inputs: title (required), message (required), course_identifier (required), published. Returns: result

create_moduleC

Create a new module (starts unpublished).

Inputs: name (required), course_identifier (required), position. Returns: result

create_pageB

Create a new wiki page. Defaults to unpublished (draft) unless published=True.

Inputs: title (required), course_identifier (required), body, published. Returns: result

create_rubricA

Create a rubric. criteria_json is a JSON array string, each item: {"description": str, "points": number, "ratings": [{"description": str, "points": number}, ...]} Example: '[{"description": "Content", "points": 30, "ratings": [{"description": "Excellent", "points": 30}, {"description": "Weak", "points": 10}]}]' Pass associate_with_assignment_id to attach it to an assignment for grading immediately.

Inputs: title (required), criteria_json (required), course_identifier (required), free_form_comments, +1 more. Returns: result

delete_announcementC

Permanently delete an announcement. This cannot be undone.

Inputs: announcement_id (required), course_identifier (required). Returns: result

delete_assignmentA

Permanently delete an assignment and its submissions/grades. This cannot be undone.

Inputs: assignment_id (required), course_identifier (required). Returns: result

delete_course_fileA

Permanently delete a course file. This cannot be undone.

Inputs: file_id (required). Returns: result

delete_discussion_topicA

Permanently delete a discussion topic (or announcement — they share the same endpoint). This cannot be undone.

Inputs: topic_id (required), course_identifier (required). Returns: result

delete_moduleA

Permanently delete a module and all its items (the items themselves, e.g. pages, are not deleted). This cannot be undone.

Inputs: module_id (required), course_identifier (required). Returns: result

delete_pageB

Permanently delete a wiki page. This cannot be undone.

Inputs: page_url (required), course_identifier (required). Returns: result

delete_peer_reviewC

Remove a previously assigned peer review.

Inputs: assignment_id (required), reviewee_user_id (required), reviewer_user_id (required), course_identifier (required). Returns: result

delete_rubricA

Permanently delete a rubric. This cannot be undone.

Inputs: rubric_id (required), course_identifier (required). Returns: result

edit_page_contentB

Replace a page's title and/or body. Only fields you pass are changed.

Inputs: page_url (required), course_identifier (required), body, title. Returns: result

get_assignment_analyticsA

Get score distribution and on-time/late/missing rates for every assignment in a course.

Inputs: course_identifier (required). Returns: result

get_assignment_detailsC

Get full details for one assignment, including description.

Inputs: assignment_id (required), course_identifier (required). Returns: result

get_conversation_detailsB

Get the full message thread for one conversation.

Inputs: conversation_id (required). Returns: result

get_course_detailsB

Get details for one course (code, name, dates, time zone, role).

Inputs: course_identifier (required). Returns: result

get_course_structureC

Get the full module/item tree for a course in one call.

Inputs: course_identifier (required). Returns: result

get_discussion_topic_detailsB

Get full details and message body for one discussion topic.

Inputs: topic_id (required), course_identifier (required). Returns: result

get_front_pageA

Get the course's front page (homepage) title and body.

Inputs: course_identifier (required). Returns: result

get_page_contentB

Get the full body of a specific wiki page by its URL slug.

Inputs: page_url (required), course_identifier (required). Returns: result

get_quiz_detailsB

Get details for one classic quiz, including question count and time limit.

Inputs: quiz_id (required), course_identifier (required). Returns: result

get_rubricA

Get a rubric's criteria, point values, and rating levels — needed to build a grade_with_rubric call.

Inputs: rubric_id (required), course_identifier (required). Returns: result

get_student_analyticsB

Get per-student engagement summary (page views, participations, on-time/late/missing) for a course.

Inputs: course_identifier (required). Returns: result

get_submission_contentA

Pull what a student actually submitted for an assignment — extracts readable text from uploaded files (PDF/DOCX/text), or returns the typed text / URL for other submission types. Also includes the submission comment thread (students often paste links there). Use this before grade_submission or grade_with_rubric so you're grading the real content.

Inputs: user_id (required), assignment_id (required), course_identifier (required). Returns: result

get_syllabusA

Get the course's Syllabus body (Settings > Course Details > Syllabus).

Inputs: course_identifier (required). Returns: result

grade_submissionA

Post a grade and/or comment for one student's submission on one assignment. Use score for points-based assignments. Use grade for other grading types: 'complete'/'incomplete' (pass_fail), a letter like 'A-' (letter_grade), or '85%' (percent). Set excuse=True to mark the assignment excused for this student instead of scoring it.

Inputs: user_id (required), assignment_id (required), course_identifier (required), grade, +3 more. Returns: result

grade_with_rubricB

Grade a submission using a rubric already associated with the assignment. criterion_scores: {criterion_id: points, ...} — get criterion_ids from get_rubric. criterion_comments: optional {criterion_id: comment text, ...}.

Inputs: user_id (required), assignment_id (required), criterion_scores (required), course_identifier (required), +2 more. Returns: result

hide_gradesB

Hide an assignment's grades from students (the inverse of post_grades).

Inputs: assignment_id (required), course_identifier (required). Returns: result

list_announcementsA

List announcements in a course, correctly distinguishing posted vs delay-scheduled.

Inputs: course_identifier (required). Returns: result

list_assignmentsB

List assignments in a course with due dates and points.

Inputs: course_identifier (required). Returns: result

list_conversationsA

List your Canvas Inbox conversations. scope: inbox, unread, starred, sent, archived.

Inputs: scope. Returns: result

list_course_filesC

List files uploaded to a course.

Inputs: course_identifier (required). Returns: result

list_coursesA

List all courses the authenticated user is enrolled in or teaches.

Inputs: no named parameters. Returns: result

list_discussion_entriesA

List top-level entries (posts) in a discussion topic, with reply counts.

Inputs: topic_id (required), course_identifier (required). Returns: result

list_discussion_topicsB

List discussion topics (excluding announcements) in a course.

Inputs: course_identifier (required). Returns: result

list_missing_submissionsA

List students with missing (past-due, not submitted) work across all assignments in a course, grouped by assignment. Use for 'who hasn't turned in X?'

Inputs: course_identifier (required). Returns: result

list_modulesA

List modules in a course with publish status and item counts.

Inputs: course_identifier (required). Returns: result

list_pagesB

List wiki pages in a course with publish status and last-updated time.

Inputs: course_identifier (required). Returns: result

list_peer_reviewsB

List peer review assignments (who's reviewing whom) for an assignment.

Inputs: assignment_id (required), course_identifier (required). Returns: result

list_quiz_submissionsC

List student submissions for a classic quiz with scores and completion status.

Inputs: quiz_id (required), course_identifier (required). Returns: result

list_quizzesA

List classic quizzes in a course with due dates and points. Does not include New Quizzes.

Inputs: course_identifier (required). Returns: result

list_rubricsC

List rubrics available in a course.

Inputs: course_identifier (required). Returns: result

list_sectionsC

List sections in a course with student counts.

Inputs: course_identifier (required). Returns: result

list_submissionsC

List submissions for an assignment with grade/submission status.

Inputs: assignment_id (required), course_identifier (required). Returns: result

list_ungraded_submissionsA

The grading queue: every submitted-but-ungraded submission in a course, grouped by assignment. Start here for 'what do I need to grade?'

Inputs: course_identifier (required). Returns: result

list_usersA

List users in a course by enrollment type (student, teacher, ta, observer, designer).

Inputs: course_identifier (required), enrollment_type. Returns: result

post_discussion_entryB

Post a new top-level entry in a discussion topic.

Inputs: message (required), topic_id (required), course_identifier (required). Returns: result

post_gradesA

Release hidden ("muted") grades on an assignment so students can see them. graded_only=True (default) posts only submissions that have been graded. Students are notified when grades post — use deliberately, not as a test.

Inputs: assignment_id (required), course_identifier (required), graded_only. Returns: result

read_course_fileA

Fetch a course file's content and return extracted text. Supports PDF, DOCX, and plain text. Find file_id via list_course_files.

Inputs: file_id (required), course_identifier (required). Returns: result

reply_to_discussion_entryC

Reply to a specific entry in a discussion topic.

Inputs: message (required), entry_id (required), topic_id (required), course_identifier (required). Returns: result

send_messageA

Send a Canvas Inbox message to one or more users. recipient_user_ids is a comma-separated list of Canvas user IDs (get IDs from list_users). Set group_conversation=True to send one message all recipients can see/reply to together; otherwise each recipient gets a separate 1:1 conversation.

Inputs: body (required), recipient_user_ids (required), subject, course_identifier, +1 more. Returns: result

update_announcementB

Update an existing announcement's title, message, and/or scheduled post date.

Inputs: announcement_id (required), course_identifier (required), title, message, +1 more. Returns: result

update_assignmentA

Update an assignment's name, description, points, due date, published state, and/or peer review settings. peer_reviews=True must be set before assign_peer_review will work on this assignment.

Inputs: assignment_id (required), course_identifier (required), name, due_at, +5 more. Returns: result

update_moduleC

Rename, publish/unpublish, or reposition a module.

Inputs: module_id (required), course_identifier (required), name, position, +1 more. Returns: result

update_module_itemB

Rename, publish/unpublish, reposition, or re-indent a module item.

Inputs: item_id (required), module_id (required), course_identifier (required), title, +3 more. Returns: result

update_page_settingsC

Publish or unpublish a page.

Inputs: page_url (required), published (required), course_identifier (required). Returns: result

upload_course_fileB

Upload a local file (from this machine) into a course's Files. folder_path is the destination folder within the course (e.g. "Week 1"), defaults to the course's root.

Inputs: local_file_path (required), course_identifier (required), folder_path. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.3/5.0

    Scored across 64 tools

    Disambiguation4/5

    Despite the large number of tools, most target distinct resources and actions (e.g., list_courses vs. list_assignments). However, a few overlaps exist: delete_discussion_topic also handles announcements, and get_course_structure overlaps with list_modules and list_pages in purpose. These are minor and descriptions clarify boundaries.

    Naming Consistency5/5

    All tools follow a consistent verb_noun pattern (list_, get_, create_, update_, delete_, etc.). Even less common verbs like post_grades, hide_grades, and send_message adhere to the same style. This makes the set predictable and easy to navigate.

    Tool Count4/5

    64 tools is high, but the server covers a wide Canvas LMS domain including courses, assignments, grades, modules, pages, discussions, files, quizzes, announcements, messages, peer reviews, and analytics. Each tool has a clear purpose, so the count feels justified rather than bloated, though it may be intense for agents to discover.

    Completeness4/5

    The surface is comprehensive with full CRUD for many resources (assignments, pages, modules, announcements, rubrics) and extensive grading/peer review capabilities. Minor gaps exist (e.g., no course editing or user creation), but core workflows like grading and course content management are fully covered.

    Related MCP Connectors

    Related MCP Servers