canvas-mcp

Canvas LMS MCP Server (canvas-mcp)

<!-- mcp-name: io.github.admin978/canvas-mcp -->

   

Ask Claude about your Canvas courses, assignments, deadlines, modules, and grades from one place.

canvas-mcp is a local-first MCP server for Canvas LMS users (students, instructors, and MCP builders). It turns Canvas REST API actions into MCP tools that work from Claude Code, Claude Desktop, and other MCP-compatible clients.

Status: alpha. Single-user, no warranty, API surface may still shift. File issues if it breaks.

Who this is for

• Students who want one view across multiple courses
• Educators who want faster access to assignments, modules, and announcements
• MCP users who want Canvas data in local Claude workflows

What you can ask Claude

• “What assignments are due this week across all my active courses?”
• “Show upcoming events and planner items for next week.”
• “List my current grades by course.”
• “Get the modules (with items) for course 12345.”
• “Show announcements for course 12345.”
• “Fetch the page syllabus from course 12345.”

Quick start

Prerequisite: Python 3.10+.

1) Create a Canvas personal access token

In Canvas: Account → Settings → Approved Integrations → + New Access Token. Copy the token shown (Canvas does not show it again later).

2) Install

From PyPI (recommended):

pip install canvas-local-mcp

Or from source:

git clone https://github.com/admin978/canvas-mcp.git && cd canvas-mcp
python3 -m venv .venv && source .venv/bin/activate
pip install -e .

3) Configure ~/.canvas.env

curl -fsSL https://raw.githubusercontent.com/admin978/canvas-mcp/main/.canvas.env.example -o ~/.canvas.env
chmod 600 ~/.canvas.env
# edit ~/.canvas.env: set CANVAS_BASE_URL (institution root, no /api/v1)
# and paste the token into CANVAS_TOKEN

For token safety guidance (least privilege, file permissions, rotation/revocation, and vulnerability reporting), see SECURITY.md.

4) Register in your MCP client

Claude Code:

claude mcp add canvas-local -- canvas-local-mcp

Claude Desktop:

• macOS example path: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows example path: %APPDATA%\Claude\claude_desktop_config.json
• Linux example path: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "canvas-local": {
      "command": "canvas-local-mcp"
    }
  }
}

Tools exposed

• list_courses
• list_assignments
• list_modules
• list_announcements
• get_page
• get_file_info
• get_grades
• planner_items
• upcoming_events
• todo

Bulk dump

canvas-local-mcp-dump downloads course files and related content for offline indexing.

canvas-local-mcp-dump              # all active courses
canvas-local-mcp-dump 12345 67890  # specific course IDs

Output goes to ./canvas-dump/ by default. Override with CANVAS_DUMP_DIR=/path/to/dir.

Local-first, privacy and token flow

• The server runs locally and uses stdio transport with your MCP client.
• Configuration is read from ~/.canvas.env (CANVAS_BASE_URL, CANVAS_TOKEN).
• Requests go from your local server to the official Canvas API endpoints.
• No external token broker is used in the request path.
• Security guidance and disclosure policy: SECURITY.md.

Demo

No demo GIF is currently committed yet. You can still verify the workflow quickly:

Try these prompts

After registering the MCP server, ask Claude:

• “List my active Canvas courses.”
• “What assignments are due this week across all my active courses?”
• “Show upcoming events and planner items for next week.”
• “List my current grades by course.”

Placeholder: add a short terminal/GIF walkthrough here in a future PR.

Development

Requires Python 3.10+.

pip install -e ".[dev]"
ruff check canvas_local_mcp tests   # lint
pytest                              # tests run against a mocked Canvas API — no token needed

CI runs lint + tests on Python 3.10–3.13 for every push and pull request.

Publishing a new release

Follow this order to publish a new version (e.g. 0.1.4) consistently across PyPI and the MCP Registry.

1. Bump the version everywhere

Update all three files to the new version string in a single PR:

| File | Field | |---|---| | pyproject.toml | [project].version | | canvas_local_mcp/__init__.py | __version__ | | server.json | root version and packages[0].version |

The CI test test_server_json_version_matches_package and the workflow validation step will fail if any of these three disagree.

2. Merge the PR into main

Wait for all CI checks to pass before merging.

3. Publish canvas-local-mcp to PyPI

The project uses Trusted Publisher (OIDC) for PyPI uploads. Trigger the PyPI publish workflow (or run python -m build && twine upload dist/* if you have credentials configured) before creating the tag, so the package is available when the MCP Registry fetches it.

4. Create and push the tag vX.Y.Z

git tag v0.1.4
git push origin v0.1.4

This triggers the Publish to MCP Registry workflow, which:

1. Validates that server.json, pyproject.toml, and the tag all declare the same version.
2. Publishes server.json to the MCP Registry via GitHub OIDC (no secrets needed).

5. Verify

• PyPI: https://pypi.org/project/canvas-local-mcp/
• MCP Registry: https://registry.modelcontextprotocol.io/?q=canvas-mcp
• GitHub Actions: https://github.com/admin978/canvas-mcp/actions

Important: never reuse or move an existing tag. If you need to re-publish after a mistake, bump to the next patch version (e.g. 0.1.4) and start from step 1.

Contributing, roadmap and support

• Issues / feature requests
• Pull requests are welcome for bug fixes and Canvas workflow improvements
• Roadmap direction currently lives in open issues and upcoming PRs

If this project helps you manage Canvas with Claude, please try the flow above and consider giving it a ⭐ so other students and educators can find it.

License

MIT — see LICENSE.

---

Built by AGENTE 404 S.L. · admin@agente404.com