The Course Hunters

Inspiration

Every semester, thousands of Texas A&M students face the same frustrating dilemma: which professor should I choose? We discovered anex.us/grades—a website that displays grade distributions based on structured queries. It was useful, but clunky. Students had to know exactly what to search for, navigate rigid input forms, and manually piece together insights from different queries. We thought: what if you could just ask questions naturally, like talking to a knowledgeable friend?

That's when we realized the power of MCP (Model Context Protocol) servers. The grade data existed—locked in PDFs and traditional web interfaces—but it wasn't truly accessible. We envisioned a world where students could ask "Who's the best professor for CSCE 221?" or "Find me high-GPA courses for next semester" and get instant, intelligent answers. Not another website. Not another form to fill out. Just natural conversation with AI, powered by comprehensive data.

What it does

The Course Hunters is an MCP server that transforms how students make course selection decisions at Texas A&M. Instead of navigating complex search forms or manually reading through hundreds of pages of grade reports, students simply ask questions in plain English through Claude Desktop:

"Who's the best professor for CSCE 221?"
"Show me easy AERO courses with high GPAs"
"Which semester should I take MEEN 315?"
"Find me small classes that could boost my GPA"

Our system queries 6,190+ course sections across 1,105 unique courses, providing instant, data-driven insights about:

Instructor performance and grade distributions
GPA trends over time
Best semesters to take specific courses
Low-risk courses with minimal fail rates
Small, high-GPA sections perfect for GPA optimization

The difference from anex.us/grades? Zero learning curve. Students don't need to understand database queries or remember specific search syntax. They just ask naturally, and our AI context engine delivers precisely what they need.

How we built it

Our technical pipeline transforms static PDFs into intelligent, queryable context:

Data Collection We downloaded official grade distribution PDFs directly from https://web-as.tamu.edu/gradereports/, covering Fall 2024, Spring 2024/2025, and Summer 2024—over 6,190 course sections worth of data.
Data Extraction & Processing
pdfplumber: Extracted tabular grade data from PDF documents using regex pattern matching
pandas: Cleaned, normalized, and structured data into CSV/JSON formats
Pre-computed aggregations at the course, instructor, and department levels for lightning-fast queries (<50ms response time)
MCP Server Development
FastMCP: Built the Model Context Protocol server with 11 specialized query functions
Implemented intelligent caching to eliminate redundant computation (95% reduction)
Created functions tailored to real student decision patterns: compare_instructors(), best_semester_to_take(), grade_trend_analysis(), gpa_boosters(), and more
Claude Desktop Integration Connected our MCP server to Claude Desktop using stdio transport, enabling natural language queries that map to our specialized functions.

The result? A three-stage pipeline—PDF → Structured Data → AI-Queryable Context—that turns inaccessible bureaucratic reports into actionable student intelligence.

Challenges we ran into

The PDF Nightmare: Grade distribution PDFs weren't designed for programmatic parsing. Inconsistent formatting, merged cells, edge cases with missing data—cleaning this mess into a reliable, structured dataset was our biggest technical hurdle. We spent hours debugging regex patterns and handling special cases to ensure data accuracy.

MCP Server Setup: Initially, getting the MCP server infrastructure working was challenging. Understanding the stdio transport protocol, configuring Claude Desktop correctly, debugging connection issues—it took significant trial and error before we had a stable foundation.

But once those two foundations were solid? Everything else clicked. Adding new query functions became straightforward. Performance optimization was a matter of smart pre-computation. The hardest part was the groundwork—after that, development accelerated dramatically.

Accomplishments that we're proud of

We're most proud of creating a tool that can both help and inspire students.

Help: Every semester, students make uninformed course decisions that impact their GPA, stress levels, and academic trajectory. Our tool democratizes access to grade intelligence that was previously buried in PDFs. Now, anyone can make data-driven decisions in seconds.

Inspire: This project shows students—many of whom are future engineers and computer scientists—what's possible when you combine AI, data, and real-world problem-solving. It's a tangible demonstration that you don't need massive infrastructure or corporate resources to build something impactful. Just curiosity, some Python libraries, and a willingness to solve actual problems.

We didn't just build a tool. We built proof that students can leverage cutting-edge AI technology to solve their own problems.

What we learned

This project was a masterclass in the power of MCP servers. Before this, many people saw grade data as "static information"—something you look up on a website, not something you converse with. MCP servers changed that paradigm entirely.

Key lessons:

MCP servers unlock static data: PDFs, CSVs, legacy systems—data that was previously "dead" can become intelligent, queryable context for AI. This has implications far beyond grade distributions.
Data parsing is harder than it looks: Real-world data is messy. Building robust parsers that handle edge cases taught us patience, debugging discipline, and the value of comprehensive testing.
Pre-computation beats real-time aggregation: By caching statistics at startup, we turned multi-second queries into 50ms responses. Smart architecture choices matter more than raw computational power.
Natural language changes accessibility: Removing the barrier of "knowing what to search for" fundamentally changes who can benefit from data. Accessibility isn't just about design—it's about interface paradigms.

We walked away not just with a finished project, but with a deeper understanding of how AI can transform data from passive archives into active intelligence.

What's next for The Course Hunters

Phase 1: Qualitative + Quantitative Fusion Our biggest planned feature is integrating RateMyProfessor.com data. Right now, we tell you which professor has the highest GPA. Soon, we'll also tell you which professors students actually enjoy. Imagine asking: "Who's a high-GPA professor for CSCE 221 that students say explains concepts clearly?"—combining hard grade data with real student experiences.

Phase 2: Predictive Insights

Personalized course recommendations based on your major, GPA goals, and schedule
Predict your likelihood of getting an A based on historical distributions
Optimal semester planning to balance difficulty and maximize GPA

Phase 3: Beyond Texas A&M The architecture is university-agnostic. We could expand to other schools, democratizing grade intelligence for millions of students nationwide.

Phase 4: Integration with Registration Systems Real-time alerts when high-performing sections open up, direct links to course registration—reducing friction from insight to action.

The Course Hunters started as a better interface for grade data. It became a demonstration of how MCP servers can transform static information into living, conversational intelligence. And for students at Texas A&M? It became the tool they didn't know they needed—until they couldn't imagine choosing courses without it.