<<<<<<< HEAD

Claude-for-Canvas-MCP

This is a custom built MCP server for Claude that connects it to Canvas, with assignment, announcement, to do viewing capabilities.

Canvas MCP Server

Connect your Canvas LMS to Claude.
Pull assignments, grades, announcements, and course data directly into your AI workflow via the Model Context Protocol.

What It Does • Quick Start • Tools • Architecture • Use Cases

What It Does

This is an MCP (Model Context Protocol) server that connects Canvas LMS to Claude. Once configured, Claude can directly access your Canvas data — assignments, due dates, grades, announcements, and course info — without you copy-pasting anything.

Why this exists: I'm a high school student who uses Claude daily and was tired of manually telling it what homework I had. So I built a bridge. Now Claude reads my Canvas directly and can plan my study sessions, break down assignments, and flag what's due.

What's MCP?

Model Context Protocol is an open standard that lets AI assistants connect to external data sources and tools. This server implements that protocol for Canvas LMS, so any MCP-compatible client (Claude Desktop, Claude Code, Cowork, etc.) can access your Canvas data.

Architecture

This MCP server runs as a Cloudflare Worker — a serverless function deployed at the edge. It exposes a single /mcp endpoint that speaks JSON-RPC 2.0 (the MCP wire protocol) over HTTP.

Claude (MCP Client)
    │
    ▼  JSON-RPC 2.0 over HTTP
Cloudflare Worker (/mcp)
    │
    ▼  REST API + Bearer Token
Canvas LMS (your-school.instructure.com)

The Worker handles MCP protocol negotiation (initialize, tools/list, tools/call), translates tool calls into Canvas REST API requests, and returns structured data back to Claude.

Why Cloudflare Workers? Always on, no server to maintain, globally distributed, and free tier covers personal usage easily. The Worker cold-starts in under 5ms.

Prerequisites

• Node.js 18+ — Download from nodejs.org (LTS version recommended)
• A Canvas LMS account — Any school using Canvas (Instructure) will work
• A free Cloudflare account — Sign up at cloudflare.com (no domain or credit card needed)

Quick Start

1. Get your Canvas API token

1. Log into your school's Canvas (e.g., https://your-school.instructure.com)
2. Go to Account → Settings
3. Scroll to Approved Integrations
4. Click + New Access Token
5. Name it something like Claude MCP and click Generate Token
6. Copy the token immediately — Canvas won't show it again

2. Clone and install

git clone https://github.com/teejballa/Claude-for-Canvas-Custom-MCP.git
cd Claude-for-Canvas-Custom-MCP
npm install

3. Configure your school's Canvas URL

Open wrangler.toml and replace the placeholder with your school's Canvas URL:

[vars]
CANVAS_URL = "https://your-school.instructure.com"

4. Log into Cloudflare and store your API key

# This opens your browser to authenticate with Cloudflare
npx wrangler login

# This prompts you to paste your Canvas API token — it's stored securely
# as a Cloudflare secret and never appears in your code
npx wrangler secret put CANVAS_API_KEY

When wrangler secret put runs, it will show Enter a secret value: — paste your Canvas token and press Enter.

5. Deploy

npm run deploy

You'll see output like:

Published canvas-mcp (0.5 sec)
  https://canvas-mcp.<your-subdomain>.workers.dev

Copy that URL — you'll need it in the next step. The /mcp endpoint is your MCP server.

6. Add to Claude

Claude Desktop (Cowork):

Go to Settings (gear icon) → MCP Servers → Add Server → enter a name like Canvas and paste your Worker URL with /mcp at the end:

https://canvas-mcp.<your-subdomain>.workers.dev/mcp

Claude Code (CLI):

Add to your MCP config file (~/.claude/claude_desktop_config.json or your project's .mcp.json):

{
  "mcpServers": {
    "canvas": {
      "type": "url",
      "url": "https://canvas-mcp.<your-subdomain>.workers.dev/mcp"
    }
  }
}

7. Test it

Ask Claude anything about your classes:

• "What assignments do I have due this week?"
• "What are my grades in each class?"
• "Are there any new announcements?"
• "Break down the assignment due tomorrow and help me plan my time."

If you get a 401 Unauthorized error, your API key is wrong — re-run npx wrangler secret put CANVAS_API_KEY with the correct token.

Tools

| Tool | Description | Parameters | |------|-------------|------------| | get_courses | List all active courses with names, codes, and IDs | — | | get_assignments | Get assignments for one course or all courses with due dates, points, and submission status | course_id (optional), include_past (optional) | | get_upcoming_assignments | Get assignments due soon across all courses, sorted by due date with urgency indicators | days_ahead (default: 14) | | get_grades | Get current letter grades and percentages for all enrolled courses | — | | get_announcements | Get recent announcements from all courses or a specific course | course_id (optional) |

Assignments include submission status (✅), urgency indicators (🔴 OVERDUE, 🔴 DUE TODAY, 🟡 DUE SOON, 🟢), point values, and course names.

Configuration

| Variable | Where | Description | |----------|-------|-------------| | CANVAS_URL | wrangler.toml | Your school's Canvas URL (e.g., https://your-school.instructure.com) | | CANVAS_API_KEY | Wrangler secret | Your Canvas API access token (stored securely, never in code) |

Use Cases

Homework planning — "What's due this week? Prioritize by difficulty and deadline."

Study session prep — "Pull my upcoming history test details and help me make a study plan."

Grade tracking — "How am I doing across all my classes? Flag anything below a B."

Assignment breakdown — "Read the rubric for my English essay and help me outline it."

Automated daily briefing — I use this MCP in a scheduled task that runs every morning at 9 AM. It scans Canvas alongside my calendar and email, then sends me a prioritized summary of what's due, what to work on, and what I'm forgetting — delivered via email and iMessage, no manual input needed.

How I Use It

This MCP is a core piece of my personal AI productivity system. It feeds into:

• A daily briefing that fires every morning — pulls Canvas assignments, calendar events, and email, then delivers a formatted summary via email and iMessage
• A homework planner that prioritizes assignments by due date, difficulty, and time available
• Study guide generation that reads assignment details and creates interactive HTML study materials with flashcards, quizzes, and concept maps
• A heartbeat system that checks project and homework status throughout the day

It runs alongside a custom Apple Mail MCP and Google Calendar integration — together they give Claude full context on school life without any manual input.

Tech Stack

| Component | Technology | |-----------|-----------| | Runtime | Cloudflare Workers | | Language | TypeScript | | Protocol | Model Context Protocol (MCP) — JSON-RPC 2.0 | | Transport | HTTP Streamable | | API | Canvas LMS REST API v1 | | Auth | Bearer token via Wrangler secrets |

Project Structure

canvas-mcp/
├── src/
│   └── index.ts          # MCP server — Canvas API calls, tool definitions, JSON-RPC dispatcher
├── package.json           # Dependencies (Wrangler, TypeScript, Cloudflare Workers types)
├── tsconfig.json          # TypeScript config (ES2022, strict mode)
├── wrangler.toml          # Cloudflare Worker config + Canvas URL
└── .gitignore

Local Development

To run the Worker locally, create a .dev.vars file in the project root with your Canvas API key (this file is gitignored and never committed):

# .dev.vars
CANVAS_API_KEY=your_canvas_api_token_here

Then:

npm install          # Install dependencies
npx wrangler dev     # Run locally at http://localhost:8787
npm run deploy       # Deploy to Cloudflare
npm run tail         # Tail production logs

The local server runs at http://localhost:8787/mcp — you can point a Claude MCP config at this URL during development.

Troubleshooting

| Problem | Fix | |---------|-----| | 401 Unauthorized | API key is wrong or expired — re-run npx wrangler secret put CANVAS_API_KEY | | Tools return empty results | Make sure you're enrolled in active courses in Canvas | | Can't log into Cloudflare | Run npx wrangler login again |

Contributing

If your school uses Canvas and you want to extend the tool set, PRs are welcome.

Ideas for extensions:

• get_submissions — pull submitted work and teacher feedback
• get_modules — navigate course modules and content pages
• get_discussions — read and post to discussion boards
• get_calendar_events — pull Canvas calendar events directly
• submit_assignment — submit work to Canvas from Claude

Built by

TJ Walsh — High school student building AI tools for productivity and education.

GitHub · LinkedIn

This project is not affiliated with Instructure (Canvas) or Anthropic.

b16f055 (initial commit, working canvas mcp with cloudflare workers)