waterloo-learn-mcp
MCP server that lets AI assistants access University of Waterloo LEARN course data (announcements, content, grades, assignments, outlines) and search past assessments in Google Drive.
README
waterloo-learn-mcp
Read-only local MCP server for University of Waterloo LEARN courses, slides, deadlines, grades, assignments, announcements, and course outlines.
It uses STDIO: ChatGPT desktop or Codex starts the Node process locally and talks to it through pipes. There is no public URL, tunnel, open port, MCP OAuth or token, Developer mode, or manually created plugin.
Connect to ChatGPT desktop (Windows)
1. Install the requirements
- ChatGPT desktop, signed in
- Node.js 20 or newer
- Git
- Codex CLI
Confirm Node and Codex are available in PowerShell:
node --version
codex --version
2. Download and set up the server
git clone https://github.com/LargoLardo/waterloo_learn_mcp.git
cd waterloo_learn_mcp
npm run setup
A browser opens. Sign in to LEARN with WatIAM, approve Duo, and wait for the
terminal to print Setup complete. Setup installs dependencies, builds the
server, saves the LEARN session to ignored auth.json, and registers the local
MCP in the configuration shared by ChatGPT desktop and Codex.
3. Enable and test it in ChatGPT desktop
- Fully quit ChatGPT desktop, including its Windows system-tray process, then reopen it.
- Start a new chat and type
/mcp. You can also check Settings > MCP servers; some app versions show the list under Plugins. - Confirm
waterloo-learn-mcpis present and enabled.Auth unsupportedis normal for a local STDIO server. - Ask: “List my Waterloo LEARN courses.”
Do not enable Developer mode or use Plugins > + to create a connection. That flow is for remote HTTPS MCP servers; this server is registered locally.
PDF slides work immediately. PowerPoint files require LibreOffice:
winget install TheDocumentFoundation.LibreOffice
Manual registration
If automatic setup reaches the login step but the MCP does not appear, run:
npm install
npx playwright install chromium
npm run build
npm run login
codex mcp add waterloo-learn-mcp -- node C:\absolute\path\to\waterloo_learn_mcp\dist\index.js
codex mcp list
Then fully restart ChatGPT desktop. Refresh an expired LEARN session at any
time with npm run login.
Authentication
There are two separate connections:
- ChatGPT → MCP: local STDIO pipes. No OAuth or bearer token is needed, so
the UI may report
Auth unsupported. - MCP → LEARN: authenticated HTTPS using the WatIAM/Duo browser session in
auth.json.
auth.json is never returned as MCP output. Requested course data is returned
to ChatGPT so it can answer you.
Tools
| Tool | Purpose |
|---|---|
list_courses |
List enrolled courses and IDs |
get_announcements |
Read announcements and attachments |
get_content |
Browse modules, topics, files, and links |
get_topic_file |
Render selected PDF/PPT/PPTX slides as images |
get_grades |
Read grades, weights, and feedback |
get_assignments |
Read assignments, submissions, and feedback |
get_upcoming |
Read upcoming events and due dates |
get_course_outline |
Read the official course outline/syllabus |
search_drive_files |
Search an optional configured Drive folder |
get_drive_file |
Render an allowed Drive file as page images |
All tools are read-only. Slide tools accept pages such as "4", "2-6", or
"2,4,7-9" and return up to 30 page images per call.
Optional configuration
Copy .env.local.example to .env.local only if you want login autofill,
custom cache paths, or Google Drive search. .env.local is ignored by Git.
For Drive search, enable the Google Drive API and set either
GOOGLE_DRIVE_API_KEY for a public/shared folder or a short-lived
GOOGLE_DRIVE_ACCESS_TOKEN with the drive.readonly scope. Limit searchable
folders with GOOGLE_DRIVE_FOLDER_IDS.
GitHub safety
Never commit auth.json, .env.local, or cache/. They are ignored, but check
before pushing:
git diff --cached
git ls-files auth.json .env.local "cache/**"
The second command must print nothing.
Development
npm test
npm run build
Disclaimer
This unofficial project is not affiliated with the University of Waterloo or D2L. Use it only with your account, respect university policies and course copyright, and verify important grades, deadlines, and policies in LEARN.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。