academic-assistant
MCP server that integrates GitHub, Brightspace, Outlook, and local files, enabling natural language queries for assignments, grades, emails, calendar events, repositories, and file operations.
README
📚 Academic Assistant
AI-powered academic assistant that connects your university tools in one chat interface. Ask questions in plain English and it pulls data from GitHub, Brightspace, Outlook, and your local files.
Built for UTRGV students — works with UTRGV Brightspace (D2L) and @utrgv.edu Office 365 email.
Table of Contents
- Quick Start
- Requirements
- Installation
- LLM Setup (Ollama)
- Module Setup
- Running the Assistant
- Daily Usage
- Troubleshooting
- Using OpenAI Instead of Ollama
- Architecture
- Available Tools
Quick Start
# 1. Clone the repo
git clone https://github.com/Miguel121699/academic-assistant.git
cd academic-assistant
# 2. Install Python dependencies
python -m venv .venv
source .venv/bin/activate
pip install -e .
# 3. Install Ollama (free local LLM)
brew install ollama
ollama serve &
ollama pull llama3.1:8b
# 4. Configure credentials
cp .env.example .env
# Edit .env with your API keys (see Module Setup below)
# 5. Run
academic-assistant
Requirements
- macOS (tested on Apple Silicon)
- Python 3.11+
- Ollama (free) OR an OpenAI API key ($)
- ~5GB disk space for the LLM model
Installation
Step 1: Clone the repository
git clone https://github.com/Miguel121699/academic-assistant.git
cd academic-assistant
Step 2: Create a virtual environment
python -m venv .venv
source .venv/bin/activate
Step 3: Install the package
pip install -e .
This installs four commands:
academic-assistant— Interactive chat agentacademic-server— Standalone MCP server (for use with Claude, Kiro, etc.)academic-oauth— Microsoft OAuth helper (personal accounts)academic-oauth-utrgv— UTRGV Outlook login (device code flow)
Step 4: Create your config file
cp .env.example .env
LLM Setup (Ollama)
Ollama runs AI models locally on your Mac — completely free, no API key needed.
Install Ollama
# On Apple Silicon Mac:
arch -arm64 brew install ollama
# On Intel Mac:
brew install ollama
Or download from: https://ollama.com/download
Start Ollama
# Start the server (runs in background)
ollama serve &
# Or run in foreground to see logs:
ollama serve
Download the model
# Recommended: Llama 3.1 8B (~5GB download, good tool-calling support)
ollama pull llama3.1:8b
Other model options:
# Smaller/faster (3.8GB, less accurate):
ollama pull llama3.2:3b
# Larger/smarter (requires 48GB+ RAM):
ollama pull llama3.1:70b
# Good at coding tasks:
ollama pull deepseek-coder-v2:16b
Verify it's working
ollama list
# Should show: llama3.1:8b
That's it! The assistant auto-detects Ollama at http://localhost:11434 and uses it.
Starting Ollama on each session
Ollama doesn't auto-start on boot. Each time you open a new terminal:
ollama serve &
Or set it to auto-start:
brew services start ollama
Module Setup
Edit your .env file for each service you want to connect:
nano .env
GitHub
What you get: List repos, view commits, create repos, push code files.
- Go to https://github.com/settings/tokens?type=beta
- Click "Generate new token"
- Name it
academic-assistant - Set expiration: 90 days (or no expiry)
- Repository access: Select "All repositories"
- Permissions → Repository permissions:
- Administration: Read and write (to create repos)
- Contents: Read and write (to push files)
- Metadata: Read-only (default)
- Click Generate token
- Copy the token (starts with
github_pat_...)
Add to .env:
GITHUB_TOKEN=github_pat_your-token-here
GITHUB_USERNAME=your-github-username
Token lasts: 90 days (or forever if you chose no expiry). No refresh needed.
Brightspace (UTRGV D2L)
What you get: List courses, view assignments/due dates, check grades, read announcements, see upcoming deadlines.
There are two authentication methods:
Method A: Manual Cookie Copy (works immediately)
- Open https://utrgv.brightspace.com in Chrome and log in
- Press Cmd+Option+I to open DevTools
- Click the Application tab
- In the left sidebar: Cookies →
https://utrgv.brightspace.com - Find and copy the Value of these two cookies:
d2lSessionVald2lSecureSessionVal
Add to .env:
BRIGHTSPACE_SESSION_VAL=paste-d2lSessionVal-here
BRIGHTSPACE_SECURE_SESSION_VAL=paste-d2lSecureSessionVal-here
⚠️ Cookies expire every ~24 hours. You'll need to repeat this daily.
Method B: Auto-Login (set and forget)
To skip daily cookie refresh, add your UTRGV credentials:
UTRGV_USERNAME=your.utrgv.username
UTRGV_PASSWORD=your-utrgv-password
The assistant will automatically log in via UTRGV Shibboleth SSO when cookies expire and refresh them in .env.
Note: If UTRGV uses Duo MFA, auto-login may require an additional approval step on your phone.
⚠️ Your password is stored in plaintext in .env. The .gitignore ensures it's never committed to git.
Outlook (@utrgv.edu)
What you get: Read/send emails, view/create calendar events — all from your @utrgv.edu account.
This uses Microsoft's device code flow — no Azure app registration needed.
One-time setup:
# Make sure you're in the project with venv active:
cd academic-assistant
source .venv/bin/activate
# Run the login command:
academic-oauth-utrgv
You'll see:
============================================================
To sign in with your UTRGV email:
1. Open: https://login.microsoft.com/device
2. Enter code: ABC123XYZ
3. Log in with your @utrgv.edu account
============================================================
Steps:
- Open https://login.microsoft.com/device in your browser
- Type the code shown in your terminal (NOT sent to email — it's displayed in the terminal)
- Sign in with your @utrgv.edu Microsoft account
- Approve the permissions
- Terminal will say "✅ Login successful! Outlook connected."
The tokens are saved to .env automatically.
Token lasts: ~90 days. Auto-refreshes silently each time you use it. Only need to redo academic-oauth-utrgv if you don't use it for 90+ days.
Local Files
What you get: Read, write, list, and search files on your computer.
No configuration needed! The first time the assistant tries to access a folder, you'll see a permission prompt:
🔒 File access requested:
Path: /Users/you/Desktop/homework.py
Allow access to /Users/you/Desktop?
(y)es / (n)o / (a)lways:
y— Allow for this session onlyn— Deny accessa— Allow permanently (saved to.env, never asks again)
To pre-approve directories without prompts, add to .env:
ALLOWED_DIRECTORIES=~/Desktop,~/Documents,~/Downloads,~/projects
Running the Assistant
# 1. Make sure Ollama is running
ollama serve &
# 2. Activate the environment
cd /path/to/academic-assistant
source .venv/bin/activate
# 3. Start chatting
academic-assistant
You'll see:
╭──────────────────────────────────────────────────────────────╮
│ 📚 Academic Assistant │
│ LLM: Ollama llama3.1:8b (local) │
│ Services: ✓ GitHub | ✓ Brightspace | ✓ Outlook (@utrgv.edu) │
│ │
│ Commands: /reset (new conversation) /quit (exit) /help │
╰──────────────────────────────────────────────────────────────╯
Commands
/help— Show all available tools/reset— Clear conversation history (start fresh)/quitor/exit— Exit the assistant
Example prompts
- "What assignments are due this week?"
- "List my courses"
- "Show my unread emails"
- "What's on my calendar tomorrow?"
- "List my GitHub repos"
- "Create a private repo called homework-6"
- "Find all PDF files on my Desktop"
- "Read the file at ~/Documents/CS3340/lab1.py"
Daily Usage
Here's what you need each day:
| Step | Command | When needed |
|---|---|---|
| Start Ollama | ollama serve & |
Every new terminal session |
| Activate env | source .venv/bin/activate |
Every new terminal session |
| Run assistant | academic-assistant |
When you want to chat |
| Refresh Brightspace cookies | See Method A | Every ~24h (unless using auto-login) |
One-liner to start everything:
cd ~/path/to/academic-assistant && ollama serve & source .venv/bin/activate && academic-assistant
Troubleshooting
"Connection refused" or Ollama not responding
# Check if Ollama is running:
curl http://localhost:11434/api/tags
# If not, start it:
ollama serve &
# Verify model is downloaded:
ollama list
Brightspace says "Session expired"
Your cookies expired. Either:
- Re-copy cookies from browser (Method A)
- Or add
UTRGV_USERNAME/UTRGV_PASSWORDto.envfor auto-refresh
Outlook says "Token refresh failed"
Re-run the device code flow:
academic-oauth-utrgv
GitHub shows "401 Unauthorized"
Your token expired or lacks permissions. Generate a new one at https://github.com/settings/tokens
LLM makes up fake data
Known issue with small local models. The assistant has guardrails to catch most hallucinations. If you see clearly fake data (fictional course names, made-up assignments), just ignore it and rephrase your question. Upgrading to a larger model or OpenAI fixes this.
Agent asks for my password in chat
Never type credentials into the chat. This is the local LLM hallucinating an auth prompt. Hit Enter to skip it or Ctrl+C and restart. The CLI has filters to catch these, but small models occasionally get through.
Using OpenAI Instead of Ollama
If you have an OpenAI API key, the assistant uses GPT-4o which is significantly better at tool-calling and never hallucinates fake data.
- Get a key at https://platform.openai.com/api-keys
- Add to
.env:OPENAI_API_KEY=sk-your-key-here - Restart
academic-assistant
The banner will show LLM: OpenAI gpt-4o and you don't need Ollama running.
Cost: ~$0.01-0.05 per conversation (very cheap for personal use).
Priority: If both are configured, OpenAI is used over Ollama.
Architecture
┌─────────────────────────────────────────────────────────┐
│ Academic Assistant │
│ │
│ ┌─────────────────────────────────────────────────┐ │
│ │ Agent Loop (OpenAI or Ollama) │ │
│ │ - Receives user message │ │
│ │ - Decides which tools to call │ │
│ │ - Feeds results back to LLM │ │
│ │ - Returns final answer │ │
│ └──────────────────────┬──────────────────────────┘ │
│ │ │
│ ┌──────────┐ ┌────────┴───┐ ┌──────────┐ ┌────────┐ │
│ │ GitHub │ │Brightspace │ │ Outlook │ │ Files │ │
│ │ Tools │ │ Tools │ │ Tools │ │ Tools │ │
│ └──────────┘ └────────────┘ └──────────┘ └────────┘ │
│ │ │ │ │ │
└───────┼──────────────┼──────────────┼────────────┼─────┘
│ │ │ │
GitHub API D2L Valence MS Graph API Local FS
(cookies) (device code)
Available Tools (21 total)
GitHub (5 tools)
| Tool | Description |
|---|---|
github_list_repos |
List your repositories |
github_get_repo |
Get repo details + recent commits |
github_create_repo |
Create a new repository |
github_push_file |
Create/update a file in a repo |
github_list_commits |
List recent commits |
Brightspace (6 tools)
| Tool | Description |
|---|---|
brightspace_list_courses |
List enrolled courses |
brightspace_list_assignments |
List assignments for a course |
brightspace_get_assignment |
Full assignment details + rubric |
brightspace_get_grades |
Current grades for a course |
brightspace_get_announcements |
Course announcements |
brightspace_get_upcoming |
All upcoming due dates across courses |
Outlook (5 tools)
| Tool | Description |
|---|---|
outlook_list_emails |
List emails (with filtering) |
outlook_get_email |
Full email content |
outlook_send_email |
Send an email |
outlook_list_events |
Upcoming calendar events |
outlook_create_event |
Create a calendar event |
Local Files (5 tools)
| Tool | Description |
|---|---|
fs_list_directory |
List files in a folder |
fs_read_file |
Read a text file |
fs_write_file |
Write/create a file |
fs_search_files |
Search files by name pattern |
fs_file_info |
Get file size, date, type |
License
MIT
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。
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。
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。
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。