ProfGraph

AI professor intelligence for any LLM. Ratings, NLP-extracted teaching style, grade distributions, and data-backed recommendations -- served via MCP and REST API.

Students ask "who should I take for CS 4348?" and get an answer grounded in 7,000+ historical student outcomes instead of hallucinated guesses.

How It Works

Student -> Claude / ChatGPT / any LLM
              |
              v
         ProfGraph (MCP + REST API)
              |
              +-- RateMyProfessors GraphQL --> ratings, tags, reviews
              +-- NLP pipeline             --> teaching style classification
              +-- Nebula Trends API        --> grade distributions (FOIA data)
              +-- Prerequisite graph       --> course dependencies
              +-- Community intel (SQLite) --> crowdsourced syllabus data

Quick Start

Local (stdio, for Claude Code / Claude Desktop)

git clone https://github.com/oneKn8/profgraph.git
cd profgraph
pip install -e .

Add to Claude Desktop MCP config (~/.config/claude/claude_desktop_config.json):

{
  "mcpServers": {
    "profgraph": {
      "command": "python3",
      "args": ["-m", "profgraph"],
      "cwd": "/path/to/profgraph"
    }
  }
}

Hosted (HTTP, for ChatGPT / remote clients)

pip install -e ".[deploy]"
PROFGRAPH_TRANSPORT=streamable-http python -m profgraph

Server starts on http://0.0.0.0:8000 with both MCP (/mcp) and REST API (/api/*).

Docker

docker build -t profgraph .
docker run -p 8000:8000 profgraph

Fly.io

fly volumes create profgraph_data --region dfw --size 1
fly launch

10 MCP Tools

Tool	Description
list_universities	Show supported schools and data availability
search_professors	Find professors by name, optionally filter by department
get_professor_profile	Full profile: ratings, NLP teaching style, tags, courses, reviews
get_grade_distribution	Historical grade data by semester (FOIA/TPIA public records)
compare_professors	Side-by-side comparison table
predict_grade	Grade probability given student GPA vs historical distribution
recommend_professor	Ranked recommendations by learning style and priorities
get_prerequisites	Course prerequisite tree with depth traversal
submit_intel	Contribute syllabus data (exam weights, curve, textbook, notes)
get_intel	Retrieve community-contributed course intel

REST API (for ChatGPT Actions)

Endpoint	Description
GET /api/universities	List supported schools
GET /api/search?university=utd&query=Smith	Professor search
GET /api/profile?university=utd&professor=Jason+Smith	Full profile
GET /api/grades?university=utd&course=CS+3341	Grade distributions
GET /api/predict?university=utd&course=CS+3341&professor=smith&gpa=3.2	Grade prediction
GET /api/prerequisites?course=CS+4348	Prerequisite tree
GET /api/openapi.json	OpenAPI 3.1 spec (import into ChatGPT)

ChatGPT setup: Create a custom GPT -> Add Action -> Import from URL -> https://yourserver/api/openapi.json

Supported Universities

University	Key	Professor Data	Grade Data
UT Dallas	utd	yes	yes
Texas A&M	tamu	yes	--
UT Austin	utaustin	yes	--
UT Arlington	uta	yes	--
U of Houston	uh	yes	--
Rice University	rice	yes	--
U of North Texas	unt	yes	--

All universities have full professor support (search, profiles, NLP teaching style, comparisons, recommendations). Grade distributions currently available for UTD via Nebula Trends API.

NLP Teaching Style

Extracted from RMP review text using keyword pattern matching:

• Exam style: straightforward, mixed, ambiguous, tricky
• Homework load: light, moderate, heavy
• Lecture quality: clear, mixed, unclear
• Curve likelihood: none, low, medium, high, guaranteed
• Accessibility: high, medium, low
• Boolean signals: uses textbook, records lectures, provides practice exams
• Warnings: extracted from low-rated reviews (exam/HW mismatch, grading issues, disorganization)
• Student types: best for / challenging for classifications

Example Output

$ profgraph search_professors utd "Jason Smith"

# Jason Smith
Computer Science | UTD

Rating: 3.3/5 | Difficulty: 4.3/5 | Would Take Again: 55%

## Teaching Style (NLP-extracted)
- Exam Style: ambiguous
- Homework Load: heavy
- Lecture Quality: clear
- Curve Likelihood: none
- Accessibility: high

## Teaching Style Tags
- Lots of homework (151x)
- Tough grader (104x)
- Accessible outside class (76x)
- Clear grading criteria (28x)
- Amazing lectures (28x)

Note: the "Teaching Style Tags" above come from RateMyProfessors' teacherRatingTags field, which has degraded upstream and now frequently returns empty. ProfGraph handles this gracefully (the tag list is simply omitted); the NLP-extracted "Teaching Style" section does not depend on it.

Data Sources

Source	Method	Auth
RateMyProfessors	GraphQL API	Public token
UTD Nebula Trends	REST API	None
Prerequisites	Hardcoded (UTD CS 2025-2026 catalog)	N/A
Community Intel	SQLite (user-contributed)	N/A

Configuration

Env Var	Default	Description
PROFGRAPH_TRANSPORT	stdio	stdio or streamable-http
PROFGRAPH_HOST	0.0.0.0	HTTP server bind address
PROFGRAPH_PORT	8000	HTTP server port
PROFGRAPH_RMP_AUTH	Basic dGVzdDp0ZXN0	RMP auth token (public, configurable)

Development

pip install -e ".[dev]"
pytest tests/ --ignore=tests/test_live.py --ignore=tests/test_e2e.py

54 unit tests (models, cache, NLP, prerequisites, intel). Live and E2E tests require network access.

License

MIT