Canvas LMS管理MCP服务

Canvas MCP Server

Connect AI assistants to Canvas LMS — manage courses, grade submissions, edit pages, and analyze rubrics through natural conversation.

Quick Start

1. Get a Canvas API token — Canvas → Account → Settings → Approved Integrations → New Access Token
2. Install via Claude Desktop Extension or npx @r-huijts/canvas-mcp
3. Try a prompt — "List all my active Canvas courses"

Table of Contents

• Features
• Prerequisites
• Installation
• Configuration
• Claude Desktop Integration
• Other MCP Clients
• Usage Examples
• Student Data Privacy
• Tool Reference
• Available Prompts
• Troubleshooting
• Performance & Caching
• Development
• Contributing
• Security Notes
• Related Documentation
• License

Features

• Courses — list active courses, post announcements
• Assignments — create, update, delete assignments and assignment groups; bulk date updates
• Submissions — grade work, post feedback, download submission files
• Rubrics — view rubrics, analyze statistics, attach rubrics to assignments
• Students — enrollment lists with privacy-first anonymization
• Sections — list sections and section-filtered submissions
• Modules — full module and module-item CRUD
• Files — browse the course Files section: folder trees, file lists, and per-folder contents
• Pages — edit content, manage revisions, and use the styleguide system (generate-styleguide, patch-page-content)
• Quizzes — full quiz, question, and question-group CRUD
• Quiz results — per-student answers, item analysis, report generation, and regrading
• ePortfolios — list and read student ePortfolios
• Prompts — analyze-rubric-statistics for multi-assignment rubric visualizations
• Performance — ETag-based response caching to reduce API load and token use

69 tools and 1 prompt in total. See docs/TOOLS.md for the full parameter reference.

Prerequisites

• Node.js v16 or higher (for npm/source installs; not required for the Desktop Extension)
• A Canvas API token with access to the courses you intend to manage
• Your Canvas instance URL (e.g. https://yourschool.instructure.com)

Note: The server defaults to https://fhict.instructure.com if CANVAS_BASE_URL is not set. Set this variable to your own institution's Canvas URL.

Installation

Option 1: Desktop Extension (Easiest)

One-click installation with Claude Desktop Extensions:

1. Download the latest extension from GitHub Releases (.mcpb or .dxt format, depending on the release)
2. Open the file with Claude Desktop (double-click or drag-and-drop)
3. Click "Install"
4. Configure your Canvas API token and base URL through the Claude Desktop UI

Benefits: no terminal required, secure token storage in the OS keychain, bundled dependencies, cross-platform (macOS and Windows).

To build the extension yourself, see DESKTOP_EXTENSION.md.

Option 2: NPM Package (Recommended)

npm install -g @r-huijts/canvas-mcp

Or run directly without installing:

npx @r-huijts/canvas-mcp

Option 3: From Source

git clone https://github.com/r-huijts/canvas-mcp
cd canvas-mcp
npm install
cp .env.example .env   # then edit with your credentials
npm run build
npm start

For development with auto-reload:

npm run dev

Configuration

Set these environment variables (via .env file, MCP client config, or shell):

See .env.example for a template.

Getting a Canvas API Token

1. Log in to your Canvas instance
2. Go to Account → Settings → Approved Integrations
3. Click + New Access Token
4. Copy the token — you won't be able to see it again

Your token must belong to a user with teacher (or equivalent) access to the courses you want to manage. Canvas personal access tokens inherit the permissions of the account that created them.

Claude Desktop Integration

1. Open Claude Desktop's configuration file:

   macOS:

   复制

   code ~/Library/Application\ Support/Claude/claude_desktop_config.json
   

   Windows:

   复制

   code %AppData%\Claude\claude_desktop_config.json
   

2. Add the Canvas MCP server:

   NPM package (recommended):

   复制

   {
     "mcpServers": {
       "canvas": {
         "command": "npx",
         "args": ["-y", "@r-huijts/canvas-mcp"],
         "env": {
           "CANVAS_API_TOKEN": "your_token_here",
           "CANVAS_BASE_URL": "https://your-canvas-instance.com"
         }
       }
     }
   }
   

   From source:

   复制

   {
     "mcpServers": {
       "canvas": {
         "command": "node",
         "args": ["/absolute/path/to/canvas-mcp/dist/index.js"],
         "env": {
           "CANVAS_API_TOKEN": "your_token_here",
           "CANVAS_BASE_URL": "https://your-canvas-instance.com"
         }
       }
     }
   }
   

3. Restart Claude Desktop

The -y flag tells npx to accept the package installation prompt automatically.

Other MCP Clients

Any MCP client that supports stdio transport can use the same configuration pattern. Replace the config file path with your client's equivalent:

{
  "mcpServers": {
    "canvas": {
      "command": "npx",
      "args": ["-y", "@r-huijts/canvas-mcp"],
      "env": {
        "CANVAS_API_TOKEN": "your_token_here",
        "CANVAS_BASE_URL": "https://your-canvas-instance.com"
      }
    }
  }
}

This works with Cursor, VS Code MCP extensions, and other stdio-based clients. Consult your client's MCP documentation for where to place the config file.

Usage Examples

Copy-paste these prompts into your AI assistant after connecting the server:

List all my active Canvas courses

Post an announcement to course 12345 titled "Week 3 Update" with a summary of this week's topics

Show rubric statistics for assignment 67890 in course 12345

List all folders and files in the Files section of course 12345

Generate a styleguide for course 12345, then patch the syllabus page to match it

List all students in course 12345 with their actual names and emails

Student Data Privacy

This server includes privacy-first anonymization for student data. By default, student names and emails are pseudonymized; you can request real identities using natural language.

Default behavior:

• Names become Student 1, Student 2, etc.
• Emails become student1@example.com, student2@example.com
• The same student always gets the same pseudonym across calls
• Teacher and admin names are never anonymized

Requesting real data:

List all students in course 123, but show their actual names and emails

Affected tools: list-students, list-assignments (with submission data), list-assignment-submissions, list-section-submissions, list-rubric-assessments, get-submission-documents

Each affected tool accepts an anonymous parameter (default: true). Your AI assistant sets anonymous: false when you ask for real names.

✅ "Excellent analysis! - Prof. Johnson"
❌ "I found this confusing - Student 1"

Tool Reference

Full parameter reference: docs/TOOLS.md

Available Prompts

analyze-rubric-statistics

Analyzes rubric statistics for formative assignments in a course and creates visualizations.

• Required: courseName (string)
• Creates grouped stacked bar and grouped bar charts across assignments and criteria
• Includes progression analysis and trend identification

Troubleshooting

Server not appearing in Claude Desktop

• Verify JSON syntax in your config file
• Use absolute paths for source installs (dist/index.js, not build/index.js)
• Ensure your Canvas API token is valid
• Restart Claude Desktop

Connection errors

• Confirm your token has access to the target courses
• Verify CANVAS_BASE_URL points to your institution's Canvas instance
• Check MCP logs:
  复制

  # macOS
  tail -f ~/Library/Logs/Claude/mcp*.log
  # Windows
  type %AppData%\Claude\Logs\mcp*.log
  

NPX issues

• On Windows, ensure npm/npx are on your PATH
• For permission errors, try running Claude Desktop as administrator (Windows)
• Corporate networks may require npm proxy configuration

Debug logging

The server logs errors to stderr. Redirect when running manually:

node dist/index.js 2> debug.log

Performance & Caching

The server caches stable Canvas API responses to avoid redundant fetches and reduce token consumption.

How it works:

• First request stores the response body and any ETag or Last-Modified header
• Subsequent requests send conditional GETs (If-None-Match / If-Modified-Since); Canvas returns 304 Not Modified when data is unchanged
• If Canvas omits validator headers, a 30-minute TTL is used as fallback
• Write operations (POST, PUT, DELETE) evict affected cache entries

What is never cached:

• Submission and grade data (/submissions endpoints) — always fetched live
• Paginated fetchAllPages() calls (students, submissions, ePortfolios) bypass the ETag cache and always hit the network — relevant for large courses

Development

Architecture

MCP Client → stdio → src/index.ts → src/tools/* → CanvasClient → Canvas REST API

Key modules:

• src/index.ts — server bootstrap and tool registration
• src/canvasClient.ts — HTTP client, ETag caching, pagination
• src/anonymizer.ts — student data pseudonymization
• src/tools/ — one file per domain, each exports register*Tools()

Tool registration

This server uses the MCP TypeScript SDK (v1.29.0). Each tool is registered with server.tool() using a five-argument form:

1. Tool name (string)
2. Tool description (string)
3. Input schema (Zod schema)
4. Tool annotations ({ readOnlyHint?, destructiveHint?, idempotentHint? })
5. Execute function

server.tool(
  "list-courses",
  "List all active courses for the authenticated user. Returns course name, ID, course code, and term.",
  {},
  { readOnlyHint: true },
  async () => {
    return {
      content: [{ type: "text", text: "..." }]
    };
  }
);

Local development

npm install
npm run build    # compile TypeScript to dist/
npm start        # run compiled server
npm run dev      # run from source with tsx (hot reload)

Contributing

1. Fork the repository
2. Create a feature branch
3. Install dependencies: npm install
4. Build: npm run build
5. Submit a pull request

There is currently no automated test suite — manual verification via an MCP client is the primary testing approach.

Security Notes

1. API token security

   • Never commit your Canvas API token to version control
   • Use environment variables or secure configuration
   • Rotate tokens periodically

2. Permissions

   • Use tokens with the minimum access needed for your use case
   • Review Canvas API access logs periodically

Related Documentation

• docs/TOOLS.md — full tool parameter reference
• DESKTOP_EXTENSION.md — building and packaging the Claude Desktop Extension
• .env.example — environment variable template

License

MIT — see LICENSE.