Canvas LMS MCP Server


An MCP (Model Context Protocol) server that connects AI assistants like Claude to your Canvas LMS
account. Query grades, assignments, missing work, and more through natural language.

Perfect for: Parents monitoring their child's academic progress, students tracking their own
work, or anyone who wants to interact with Canvas data through AI.

What Can It Do?

Ask Claude things like:

- "What assignments are due this week?"
- "Show me missing assignments across all classes"
- "What's my grade in Biology?"
- "Are there any late assignments I should know about?"
- "What's on the to-do list for the next 7 days?"

Installation Options

Choose the method that works best for you:

| Method | Best For | Credential Storage |
| ------------------------------------------------------------ | ------------------------------ | ------------------------------- |
| Desktop Extension | Easy one-click install | OS Keychain (secure) |
| Claude Code (CLI) | Developers using Claude Code | Shell env vars or config file |
| npm Package | Claude Desktop + 1Password | Config file or password manager |
| Standalone CLI | Direct terminal use without AI | Environment variables |

---

Prerequisites (All Methods)

$3

1. Log into your Canvas LMS instance (e.g., https://yourschool.instructure.com)
2. Go to Account → Settings
3. Scroll to Approved Integrations
4. Click + New Access Token
5. Give it a name (e.g., "Claude MCP") and click Generate Token
6. Copy the token immediately - you won't be able to see it again

$3

Your Canvas base URL is the main URL you use to access Canvas:

- https://yourschool.instructure.com
- https://canvas.yourdistrict.org

---

Option 1: Desktop Extension (Recommended)

The easiest way to get started. Your API token is stored securely in your operating system's
keychain.

$3

1. Download the latest .mcpb file from
Releases
2. Double-click to install, or drag into Claude Desktop
3. Claude Desktop will prompt you to enter your Canvas credentials
4. Done! Start chatting about your courses

$3

- API Token → macOS Keychain / Windows Credential Manager (encrypted)
- Base URL → Claude Desktop settings
- Student ID → Claude Desktop settings (if using observer account)

---

Option 2: Claude Code (CLI)

For developers using Claude Code, the AI-powered
CLI tool.

Requires: Node.js 18+

$3

1. Add environment variables to your shell profile (~/.zshrc or ~/.bashrc):

``bash
export CANVAS_API_TOKEN="your_token_here"
export CANVAS_BASE_URL="https://yourschool.instructure.com"


Optional for parent/observer accounts:

export CANVAS_STUDENT_ID="123456"

`

2. Reload your shell:

`bash
source ~/.zshrc # or source ~/.bashrc
`

3. Add the MCP server to your Claude Code settings.

Edit ~/.claude/settings.json (global) or .mcp.json (project-level):

`json
{
"mcpServers": {
"canvas": {
"command": "npx",
"args": ["-y", "@mtgibbs/canvas-lms-mcp"],
"env": {
"CANVAS_API_TOKEN": "${CANVAS_API_TOKEN}",
"CANVAS_BASE_URL": "${CANVAS_BASE_URL}"
}
}
}
}
`

4. Restart Claude Code - the Canvas MCP will be available.

$3

- npx -y automatically downloads the package from npm on first run (no npm install needed)
- The ${VAR} syntax pulls values from your shell environment
- Claude Code will have access to all Canvas tools when chatting

$3

If you prefer to install globally:

`bash
npm install -g @mtgibbs/canvas-lms-mcp
`

Then update your config:

`json
{
"mcpServers": {
"canvas": {
"command": "canvas-lms-mcp",
"args": [],
"env": {
"CANVAS_API_TOKEN": "${CANVAS_API_TOKEN}",
"CANVAS_BASE_URL": "${CANVAS_BASE_URL}"
}
}
}
}
`

---

Option 3: npm Package

For Claude Desktop users who want more flexibility than the Desktop Extension.

Requires: Node.js 18+

$3

Edit your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows:
%APPDATA%\Claude\claude_desktop_config.json

`json
{
"mcpServers": {
"canvas": {
"command": "npx",
"args": ["-y", "@mtgibbs/canvas-lms-mcp"],
"env": {
"CANVAS_API_TOKEN": "your_token_here",
"CANVAS_BASE_URL": "https://yourschool.instructure.com"
}
}
}
}
`

Restart Claude Desktop after saving.

---

Securing Your API Token

The basic setup stores your token in plain text. Here are more secure alternatives:

$3

If you use 1Password, you can inject secrets at runtime so they never
touch your config file.

1. Install the 1Password CLI: Download here

2. Store your Canvas credentials in 1Password (note the vault and item names)

3. Create a wrapper script at ~/.config/canvas-mcp-wrapper.sh:

`bash
#!/bin/bash


Canvas MCP 1Password Wrapper

Ensure 1Password CLI is authenticated

if ! op account get &>/dev/null 2>&1; then
eval $(op signin)
fi

Export credentials from 1Password and run the MCP server

export CANVAS_API_TOKEN=$(op read "op://Private/Canvas LMS/token")
export CANVAS_BASE_URL=$(op read "op://Private/Canvas LMS/url")

Uncomment for observer accounts:

export CANVAS_STUDENT_ID=$(op read "op://Private/Canvas LMS/student_id")

exec npx -y @mtgibbs/canvas-lms-mcp
`

4. Make it executable:

`bash
chmod +x ~/.config/canvas-mcp-wrapper.sh
`

5. Update your Claude Desktop config:

`json
{
"mcpServers": {
"canvas": {
"command": "/Users/YOUR_USERNAME/.config/canvas-mcp-wrapper.sh",
"args": []
}
}
}
`

Now your token lives in 1Password and is only loaded when the MCP server starts.

$3

If you must store the token in the config file, at least restrict access:

`bash


macOS/Linux


chmod 600 ~/Library/Application\ Support/Claude/claude_desktop_config.json
`

---

For Parent/Observer Accounts

If you're a parent with an observer account linked to your child's Canvas account, add the student
ID:

$3

Enter the Student ID when prompted during setup.

$3

Add CANVAS_STUDENT_ID to your environment variables:

`json
{
"env": {
"CANVAS_API_TOKEN": "...",
"CANVAS_BASE_URL": "...",
"CANVAS_STUDENT_ID": "123456"
}
}
`

$3

1. Log into Canvas with your observer account
2. The student ID appears in URLs when viewing their information (e.g., /users/123456)
3. Or ask your school's Canvas administrator

---

Available Tools

Once connected, Claude has access to these capabilities:

| Tool | Description |
| -------------------------- | -------------------------------------------------------------- |
| get_courses | List all courses with current grades |
| get_comprehensive_status | Full academic overview in one call (grades, missing, upcoming) |
| get_missing_assignments | Assignments flagged as missing by Canvas |
| get_unsubmitted_past_due | Past-due work not yet submitted |
| get_recent_grades | Recently graded assignments with scores |
| get_upcoming_assignments | Assignments due soon for a specific course |
| get_due_this_week | All assignments due across all courses |
| list_assignments | Search/filter assignments by status |
| get_stats | Late/missing statistics by course |
| get_todo | Planner items and to-do list |

---

Standalone CLI

This package also includes a standalone command-line tool for querying Canvas directly from your
terminal—no AI required.

$3

`bash


Option 1: Install globally

npm install -g @mtgibbs/canvas-lms-mcp

Option 2: Run directly with npx (slower startup)

npx @mtgibbs/canvas-lms-mcp courses
`

$3

Set environment variables in your shell profile (~/.zshrc or ~/.bashrc):

`bash
export CANVAS_API_TOKEN="your_token_here"
export CANVAS_BASE_URL="https://yourschool.instructure.com"
export CANVAS_STUDENT_ID="123456" # Optional: for observer accounts
`

$3

`bash


List all courses with grades

canvas courses
canvas courses --format table

Get comprehensive academic status

canvas status
canvas status --format table

List missing assignments

canvas missing
canvas missing --summary # Counts by course
canvas missing --include-unsubmitted # Include past-due not yet flagged

List assignments due soon

canvas due --days 7
canvas due --show-graded # Include already-graded items

List unsubmitted past-due assignments

canvas unsubmitted
canvas unsubmitted --course-id 12345

List assignments with filters

canvas assignments --all-courses --due-this-week
canvas assignments --course-id 12345 --upcoming 7
canvas assignments --course-id 12345 --bucket overdue

List grades/submissions

canvas grades --all-courses
canvas grades --all-courses --below B
canvas grades --course-id 12345 --days 14

Show upcoming events

canvas upcoming --days 14
canvas upcoming --type assignment

Show to-do list

canvas todo --days 7
canvas todo --hide-submitted

Show late/missing statistics

canvas stats
canvas stats --hide-empty
`

$3

These options work with all commands:

| Option | Description |
| ------------------- | ---------------------------------------------------------------------------------- |
| --format | Output format: json (default) or table |
| --student | Student ID for observer accounts (uses CANVAS_STUDENT_ID from config if omitted) |
| --help | Show help for a command |

$3

- JSON (default): Machine-readable, great for scripting or piping to jq
- Table: Human-readable formatted output

`bash


JSON output (default)

canvas courses

Table output

canvas courses --format table

Pipe JSON to jq for filtering

canvas missing | jq '.[] | select(.points_possible > 50)'
`

---

Available Prompts

Prompts are pre-built conversation starters that guide Claude through common workflows:

| Prompt | Description |
| -------------------- | -------------------------------------------------------------------------- |
| daily-checkin | Quick daily overview of grades, due items, and concerns |
| week-planning | Plan the upcoming week with prioritized assignments |
| course-analysis | Deep dive into a specific course's performance |
| grade-recovery | Find opportunities to improve grades through extra credit or resubmissions |
| missing-work-audit | Comprehensive audit of all missing and late work |

$3

In Claude Desktop, you can trigger prompts directly:

1. Click the prompt icon (📝) in the chat input
2. Select a prompt from the list
3. Fill in any required parameters (like student ID)
4. Claude will execute the workflow using the appropriate tools

Or just ask naturally - Claude will use the right tools:

> "Can you do a daily check-in for my student?"

---

Example Conversations

These examples show realistic interactions, including which tools Claude calls and what output to
expect.

$3

User prompt:

> "Give me a quick update on my kid's school situation"

Tools called:

1. get_courses - retrieves all courses with current grades
2. get_missing_assignments - checks for missing work
3. get_due_this_week - finds upcoming deadlines

Claude's response:

> Here's today's overview for your student:
>
> Current Grades:
>
> - AP Biology: 92% (A-)
> - English 11: 87% (B+)
> - US History: 78% (C+)
> - Algebra 2: 91% (A-)
>
> Needs Attention:
>
> - "Chapter 12 Reading Quiz" in US History - due yesterday, not submitted
> - "Lab Report: Photosynthesis" in AP Biology - due 2 days ago
>
> Coming Up This Week:
>
> - Essay Draft (English 11) - due Thursday
> - Unit Test (Algebra 2) - due Friday
>
> I'd recommend prioritizing those two missing assignments first, especially the History quiz since
> that class has the lowest grade.

---

$3

User prompt:

> "What assignments are missing across all classes?"

Tools called:

1. get_missing_assignments - gets Canvas-flagged missing items
2. get_unsubmitted_past_due - catches items Canvas may have missed

Claude's response:

> I found 4 missing assignments:
>
> | Assignment | Course | Due Date | Points |
> | -------------------------- | ---------- | -------- | ------ |
> | Chapter 12 Reading Quiz | US History | Jan 18 | 20 |
> | Lab Report: Photosynthesis | AP Biology | Jan 16 | 50 |
> | Vocabulary Set 8 | English 11 | Jan 15 | 15 |
> | Problem Set 7.3 | Algebra 2 | Jan 10 | 25 |
>
> Total points at risk: 110 points
>
> The Biology lab report is worth the most - I'd start there. The Algebra problem set is oldest, so
> check if late submissions are still accepted.

---

$3

User prompt:

> "What's due in the next 7 days? Help me prioritize."

Tools called:

1. get_due_this_week - retrieves all assignments due in the next 7 days
2. get_stats - gets late/missing statistics for context

Claude's response:

> Here's what's coming up in the next 7 days:
>
> High Priority (worth 50+ points):
>
> - Unit 5 Test (Algebra 2) - Friday, Jan 24 - 100 points
> - Research Paper Draft (English 11) - Thursday, Jan 23 - 75 points
>
> Medium Priority:
>
> - Lab Questions (AP Biology) - Wednesday, Jan 22 - 30 points
> - Map Quiz (US History) - Tuesday, Jan 21 - 25 points
>
> Lower Priority:
>
> - Discussion Post (English 11) - Sunday, Jan 26 - 10 points
>
> Suggested Schedule:
>
> - Monday: Start research paper outline
> - Tuesday: Map quiz prep + continue paper
> - Wednesday: Bio lab questions
> - Thursday: Finish and submit paper
> - Friday: Algebra test prep
>
> Note: US History currently has 2 missing assignments, so staying current there is important.

---

Troubleshooting

$3

Your API token isn't being passed correctly. Double-check:

- The token is in the env section of your config
- There are no extra spaces or quotes around the token
- The config file is valid JSON
- If using 1Password wrapper, run it manually to test: ~/.config/canvas-mcp-wrapper.sh

$3

Make sure your Canvas URL is set and includes https://.

$3

- Restart Claude Desktop completely
- Check that your config file is valid JSON (use a JSON validator)
- Ensure Node.js 18+ is installed: node --version

$3

- Your API token may have expired - generate a new one
- Make sure the token has the necessary permissions
- Verify the base URL is correct (no trailing slash)

$3

- Verify CANVAS_STUDENT_ID is set correctly
- Confirm your observer account is properly linked to the student in Canvas
- Try the student ID without quotes (as a number)

$3

- Run op signin manually first to authenticate
- Check that the item path matches your vault: op read "op://VaultName/ItemName/field"
- Ensure the wrapper script is executable: chmod +x ~/.config/canvas-mcp-wrapper.sh

---

Privacy & Security

$3

This MCP server collects and processes the following data locally on your machine:

| Data Type | Source | Purpose |
| ------------------- | ------------- | ----------------------------------------- |
| Course information | Canvas API | Display grades and course lists |
| Assignment data | Canvas API | Show due dates, missing work, submissions |
| Student identifiers | Canvas API | Associate data with correct student |
| API credentials | User-provided | Authenticate with Canvas |

$3

- Local processing only: All data flows directly between Canvas and Claude on your machine
- No external transmission: This server does not send data to any third-party services
- No telemetry: We do not collect usage statistics, analytics, or crash reports
- No data storage: Data is fetched on-demand and not persisted between sessions

$3

This extension connects to:

- Canvas LMS (your school's instance): To retrieve academic data using your API token
- Claude (Anthropic): The AI assistant that processes your queries locally

We do not share your data with any other third parties.

$3

- Session-based: Data is only held in memory during active use
- No caching: Academic data is not cached or stored locally
- Credentials: Stored according to your installation method:
- Desktop Extension: OS Keychain (macOS Keychain / Windows Credential Manager)
- npm + 1Password: Your 1Password vault
- npm with env vars: Your config file (recommend restricting file permissions)

$3

- You can revoke access at any time by deleting your Canvas API token
- Uninstalling the extension removes all local configuration
- Your school's Canvas instance retains its own data per their policies

$3

- Token scope: Your API token only accesses data you can already see in Canvas
- Read-only operations: This server only reads data; it cannot modify assignments or grades
- No network exposure: The server only communicates via local stdio, not over the network

$3

For privacy questions or concerns:

- GitHub Issues:
github.com/mtgibbs/canvas-lms-mcp/issues
- Author: @mtgibbs

---

Support

Need help? Here's how to get support:

- Bug Reports:
Open an issue
- Feature Requests:
Open an issue
- Questions: Start a discussion
- Security Issues: Please report security vulnerabilities privately via
GitHub Security Advisories

Response times: We aim to respond to issues within a few days. This is a community project
maintained in spare time.

---

Development

This project is built with Deno and uses
dnt for npm packaging.

`bash


Clone the repo

git clone https://github.com/mtgibbs/canvas-lms-mcp.git
cd canvas-lms-mcp

Run MCP server in development (requires Deno)

deno run -A agent.ts

Run CLI in development

deno task dev courses --format table

Build npm package

deno task build:npm

Build Desktop Extension

deno task build:extension

Type check

deno task check
`

$3

Releases are fully automated via Release Please +
npm Trusted Publishing (OIDC). No tokens required!

How it works:

`
Push commits with conventional messages (feat:, fix:, etc.)
↓
Release Please creates/updates a Release PR
↓
Merge the PR when ready
↓
Release Please creates GitHub Release + tag
↓
Publish workflow runs automatically
↓
npm package published + .mcpb attached to release
`

Commit message format:

- feat: add new tool → minor version bump (0.1.0 → 0.2.0)
- fix: correct API error → patch version bump (0.1.0 → 0.1.1)
- feat!: breaking change → major version bump (0.1.0 → 1.0.0)
- chore: update deps → no release

First-time setup:

1. Publish the first version manually:
`bash
deno task build:npm
cd npm && npm publish --access public
`

2. Configure Trusted Publisher on npmjs.com:
- Go to your package → Settings → Trusted Publisher
- Select GitHub Actions
- Organization/user: mtgibbs
- Repository: canvas-lms-mcp
- Workflow filename: publish.yml`
- Click Set up connection

After that, just push commits and merge Release PRs - everything else is automatic!

---

License

MIT License - see LICENSE for details.

---

Links

- Canvas LMS API Documentation
- MCP Protocol Specification
- Claude Desktop
- 1Password CLI
- Report Issues