canvas-control

canvas-control turns Canvas LMS into a fast, scriptable download pipeline and exposes it to AI assistants via MCP.

• Download files from multiple Canvas content sources in one command

• Sync course files like git pull — skip unchanged, re-download on demand

• Check grades across all courses from the terminal or export to CSV/JSON

• Submit assignments from the terminal (file upload, text entry, URL)

• Talk to Canvas through AI assistants (Claude Desktop, Cursor, Claude Code) — search files, download by name or type, sync courses, and set paths in natural language

Star History

Related MCP server: Canvas MCP Server

Table of contents

• Why canvas-control

• Quick start

• MCP server (AI assistants)

• CLI highlights

• Documentation

• Dev Testing

• Contributing

Why canvas-control

• One command downloads from files, assignments, discussions, pages, and modules.

• Built for repeatable workflows: predictable output paths, machine-readable manifests, idempotent re-runs.

• Guided mode (download interactive) for beginners, fully scripted mode (download run) for power users.

• Grades accessible from the terminal: summaries, per-assignment breakdowns, CSV/JSON export.

• MCP server lets AI assistants query courses, grades, assignments, and sync files with natural language.

Quick start

canvas-control uses uv for environment and dependency management.

1. Create and activate a virtual environment (Python 3.12+):

uv venv --python 3.12
source .venv/bin/activate

2. Install the project and dev dependencies:

uv pip install -e '.[dev]'

3. Set your Canvas base URL once:

cvsctl config set-base-url https://your-school.instructure.com

4. Set CANVAS_TOKEN (see configuration docs) or let cvsctl prompt for it.

5. Run the onboarding wizard to configure everything interactively, or jump straight to your first command:

# Guided setup (recommended for first-time users)
cvsctl onboard

# Or go straight to listing your courses
cvsctl courses list

MCP server (AI assistants)

canvas-control includes an MCP (Model Context Protocol) server so you can interact with Canvas through AI assistants using natural language.

Available tools

Natural language file downloads

Three new tools enable a conversational download workflow directly from Claude Desktop or Cursor:

**search_course_files** — search before you download. Filter by name substring, file extension, or folder path. The assistant shows you a preview of matching files so you can confirm before anything hits disk.

**download_selected_files** — download a batch of specific files (by Canvas file ID) to any local directory. Skips files that already exist, so it's safe to re-run. Returns a per-file status: downloaded, skipped, or failed.

**set_download_path** — persist a directory path to config so you don't have to specify it every time. Set a global default or a per-course path.

**sync_course_files** — now accepts an optional destination parameter, so you can tell the assistant exactly where to put everything without touching config first.

Example conversation:

You: Search for PDF files in my Biology course
Claude: Found 8 PDFs — Lecture 1.pdf, Lecture 2.pdf, Syllabus.pdf …

You: Download those to ~/Documents/Bio
Claude: Downloaded 7, skipped 1 (already exists). Saved to ~/Documents/Bio.

You: Save that as my default path for Biology
Claude: Saved ~/Documents/Bio as the download path for Biology (course 12345).

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "canvas": {
      "command": "uv",
      "args": ["--directory", "/path/to/canvas-control", "run", "cvsctl", "mcp", "serve"],
      "env": {
        "CANVAS_TOKEN": "your-token",
        "CANVAS_BASE_URL": "https://your-school.instructure.com",
        "CANVAS_TIMEZONE": "America/Los_Angeles"
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json in your project or ~/.cursor/mcp.json globally:

{
  "mcpServers": {
    "canvas": {
      "command": "uv",
      "args": ["--directory", "/path/to/canvas-control", "run", "cvsctl", "mcp", "serve"],
      "env": {
        "CANVAS_TOKEN": "your-token",
        "CANVAS_BASE_URL": "https://your-school.instructure.com",
        "CANVAS_TIMEZONE": "America/Los_Angeles"
      }
    }
  }
}

Claude Code

claude mcp add canvas -- uv --directory /path/to/canvas-control run cvsctl mcp serve

Set CANVAS_TOKEN, CANVAS_BASE_URL, and optionally CANVAS_TIMEZONE in your shell profile or .envrc.

Example prompts

Courses, grades, and assignments:

• "What classes am I taking?"

• "What assignments are due this week?"

• "Show my grades for Biology"

• "Are there any new announcements?"

• "Mark Homework 2 complete"

Downloading files:

• "Search for slides in my CS101 course"

• "Download the lecture PDFs from my Biology course to ~/Documents/Bio"

• "Save ~/Documents/Bio as my default download path for Biology"

• "Sync all files from my Time Series course to ~/Documents/School"

• "Re-download all files for Biology, overwriting existing ones"

CLI highlights

Download all files for a course:

cvsctl download run --course 12345

View grades at a glance:

cvsctl grades summary

Export grades to CSV:

cvsctl grades export --detailed

Guided interactive download:

cvsctl download interactive

Submit an assignment:

cvsctl assignments submit --course 12345 --assignment "Homework 1" --file ./solution.py

See CLI Reference for the full command tree and all options.

Documentation

• Configuration — token setup, download paths, per-course paths

• CLI Reference — full command tree and behavior notes

• Guides — workflows, grades, interactive mode, troubleshooting

Dev Testing

When changes are merged into the main branch, update your local clone and reinstall so the MCP server picks up the latest code.

1. Pull the latest changes:

git pull origin main

2. Reinstall the package (picks up new dependencies or entry-point changes):

uv pip install -e '.[dev]'

3. Run the test suite to verify nothing is broken:

uv run pytest

4. Restart the MCP server in your AI client so it spawns a fresh process:

• Claude Desktop: quit and relaunch the app

• Cursor: open Settings → MCP and click the refresh/reconnect button

• Claude Code: the server is spawned per-session; start a new session

Contributing

1. Create a branch for your change.

2. Implement focused changes and tests.

3. Run tests:

uv run pytest

4. Optional live smoke test (requires real Canvas credentials):

export CANVAS_BASE_URL="https://your-school.instructure.com"
export CANVAS_TOKEN="your-token"
export CANVAS_TEST_COURSE_ID="12345"
uv run pytest -m live

5. Open a PR with what changed, why, and test coverage.