MCP 服务器

courseforge-mcp

Builds entire Canvas courses from a syllabus, generating IMS Common Cartridge files and importing them via the Canvas content-migrations API.

README

CourseForge turns a syllabus into a real Canvas course. Hand your AI assistant a course outline and get back modules, pages, assignments, quizzes with questions, and discussions, imported straight into Canvas LMS or packaged as a standards-compliant .imscc file that any Canvas instance can import.

Why CourseForge?

Existing Canvas MCP servers read courses well, and some create individual pages or assignments. CourseForge builds the whole course: it generates a Canvas-flavored IMS Common Cartridge (the same format Canvas itself exports) and pushes it through Canvas's content-migrations API, so one import call creates every module, page, assignment, and quiz. The content arrives unpublished, and you review it in Canvas before students see anything.

Quickstart

1 · MCP server: let your AI build courses in Canvas

npx -y courseforge-mcp init        # interactive: writes config for Claude Code/Desktop/Cursor

Then ask your assistant: “Here's my syllabus. Build it into Canvas course 12345.”

<details> <summary>Manual config / Claude Code one-liner</summary>

claude mcp add courseforge \
  -e CANVAS_BASE_URL=https://youruni.instructure.com \
  -e CANVAS_API_TOKEN=your-token \
  -- npx -y courseforge-mcp

Get a token in Canvas: Account → Settings → + New access token. </details>

2 · No API token? Build an importable file instead

npx -y @courseforge/imscc build examples/intro-to-chemistry.json

Upload the resulting .imscc in Canvas under Settings → Import Course Content → Common Cartridge 1.x Package.

3 · Agent skills for Claude Code, Codex, Cursor & 40+ agents

npx skills add jasp-nerd/courseforge      # canvas-course-builder, imscc-editor, canvas-course-qc

What you get


🏗️ `build_course_from_spec`	One tool call: CourseSpec JSON → validated `.imscc` → imported into Canvas with progress and issue reporting
📦 `@courseforge/imscc`	Build, parse, modify, and validate Canvas cartridges (the export → edit → re-import loop), with QTI 1.2 quiz generation for 8 question types
🔌 `courseforge-mcp`	20 tools: course/module/page/assignment/quiz/discussion creation, import/export, migration progress, and a pre-publish QC linter
🧠 Agent skills	Interview the teacher, apply university branding, preview the build plan, create everything unpublished. Portable across the open Agent Skills standard
🌐 `@courseforge/canvas-client`	Standalone Canvas REST client: token or browser-cookie auth, Link-header pagination, 429 retry, the 3-step upload dance
🛡️ Tested	61 tests: golden-file manifests, round-trips on real Canvas exports, zip-slip/zip-bomb/XXE guards, MSW-mocked API, MCP protocol integration

How it works

flowchart LR
    A["📄 Syllabus / course idea"] --> B["🤖 AI agent<br/>+ canvas-course-builder skill"]
    B --> C["CourseSpec JSON<br/>Zod-validated"]
    C -->|"@courseforge/imscc"| D["📦 .imscc<br/>Canvas cartridge"]
    D -->|"content_migrations API"| E["🎓 Canvas course<br/>unpublished, ready for review"]
    D -->|"manual upload"| E
    C -->|"mode: api"| E
    E -->|"export"| D

A plain JSON course description called the CourseSpec connects everything. Agents write it, the imscc package turns it into a cartridge with the same structure Canvas's own exports use (including the canvas_export.txt flag and cccv1p0 extension XMLs, which preserve module structure and assignment settings on import), and the MCP server pushes it through the same import pipeline as the Canvas UI.

Compared to existing Canvas MCP servers

	CourseForge	vishalsachdev/canvas-mcp	bruchris/canvas-lms-mcp
Build a whole course in one call	✅	—	—
Generate + import `.imscc` packages	✅	—	—
Modify existing Canvas exports	✅	—	—
Quiz generation incl. questions	✅ QTI 1.2	—	New Quizzes API
Course/content creation tools	✅	partial	✅
Grading / student-facing tools	—	✅	✅
Language	TypeScript	Python	TypeScript

vishalsachdev's and bruchris's servers cover course management (grading, feedback, student workflows) far more deeply than CourseForge does. CourseForge covers creation. They run side by side without conflict.

Repository layout

packages/shared          CourseSpec Zod schemas (the shared course format)
packages/imscc           .imscc build / parse / modify / validate + CLI
packages/canvas-client   Canvas REST client (zero MCP dependencies)
packages/mcp             The courseforge-mcp server
skills/                  Agent Skills (open SKILL.md standard)
fixtures/                Real .imscc exports used in round-trip tests
docs/                    IMSCC format ground truth, specs, launch playbook