Canvas MCP

Read-only FastMCP server for Canvas LMS study data.

This project is configured for CIAM Canvas by default:

https://ciam.instructure.com

Configuration

Set these environment variables before starting the server:

export CANVAS_BASE_URL="https://ciam.instructure.com"
export CANVAS_API_TOKEN="your-canvas-token"

Do not commit real Canvas tokens.

Related MCP server: CoursePack Local MCP Server

Install

cd /Users/zihanwang/Documents/Codex/canvas-mcp
uv sync --extra dev

Run

cd /Users/zihanwang/Documents/Codex/canvas-mcp
uv run canvas-mcp

The server speaks MCP over stdio.

MCP Client Config

Use this shape in an MCP client config. Replace your-canvas-token with your token in your local config only.

{
  "mcpServers": {
    "canvas": {
      "command": "uv",
      "args": [
        "--directory",
        "/Users/zihanwang/Documents/Codex/canvas-mcp",
        "run",
        "canvas-mcp"
      ],
      "env": {
        "CANVAS_BASE_URL": "https://ciam.instructure.com",
        "CANVAS_API_TOKEN": "your-canvas-token"
      }
    }
  }
}

Tools

• canvas_self: current Canvas user.

• canvas_courses: visible courses, defaulting to active enrollments.

• canvas_course_assignments: assignments for a course.

• canvas_upcoming_events: upcoming calendar events.

• canvas_todo: todo items for the current user.

• canvas_course_modules: modules and module items for a course.

• canvas_course_pages: wiki pages for a course, with optional search.

• canvas_course_files: files for a course, with optional folder filtering.

• canvas_submissions: assignment submissions or visible course submissions.

• canvas_raw_get: read-only escape hatch for /api/v1/... Canvas paths.

Local Verification

cd /Users/zihanwang/Documents/Codex/canvas-mcp
uv run pytest -v
uv run python -c "from canvas_mcp.server import mcp; print(mcp.name)"

Expected output from the smoke test:

Canvas LMS

Live API Smoke Test

After setting CANVAS_API_TOKEN, this should return your Canvas user profile:

cd /Users/zihanwang/Documents/Codex/canvas-mcp
uv run python - <<'PY'
from canvas_mcp.client import CanvasClient, CanvasConfig

client = CanvasClient(CanvasConfig.from_env())
try:
    print(client.get("users/self", paginate=False))
finally:
    client.close()
PY