NBA GM Simulator

Home page
Team dashboard
Trade proposal team selector
Trade proposal
Claude would *not* make the Luka trade
Chat interface

Inspiration

After the Luka Dončić trade happened, I thought to myself that even an AI agent or a random person could do a better job of being the general manager of an NBA team than Nico Harrison. So why not put that to the test?

I also wanted to build a fun, interactive application that demonstrates the power of AI assistants with external tools. The NBA trade deadline is always exciting, and I wanted to recreate that experience in a simulator that uses real NBA data, MCP, and Claude's reasoning capabilities to evaluate trades realistically.

What it does

NBA GM Simulator allows the user to take on the role of general manager of 1 of the 30 NBA teams and negotiate trades with 29 AI agents, each of which manages another NBA team. Simultaneously, these 29 agents are negotiating trades with each other. Each of these agents uses MCP to view their current players and their NBA stats through the NBA API, allowing them to make more informed decisions.

Basically, you can:

Browse current team rosters with up-to-date statistics and salary information
Propose trades with other teams in the league
Chat and negotiate with AI-controlled GMs who can accept, reject, or counter offers
Simulate league-wide trade activity to see what other teams are doing

How I built it

The application consists of three main components:

Frontend: A React/Next.js application with Tailwind CSS that provides a modern, responsive interface for users to interact with their team and propose trades.
Backend Server: A Flask server that manages the simulation state, handles trade proposals, and communicates with the AI agents for trade evaluations.
AI Integration:
- Created a Model Context Protocol (MCP) server that gives Claude 3.5 Sonnet clients access to NBA player and team data
- Implemented custom tools for player analysis, team evaluation, and trade assessment, taking additional factors like salary cap into account

Challenges I ran into

MCP Integration: This was my first time using MCP—I learned it on my flight here (literally learning on the fly). Initially I struggled with the correct approach to integrate MCP with Claude.
UI/UX Design: Presenting complex NBA data and trade mechanics in an intuitive interface
Realistic Trade Logic: Creating a system that proposes and evaluates trades similar to how real NBA GMs would think

Accomplishments I'm proud of

Successfully implementing a complete trade proposal and evaluation system that considers multiple factors
Creating an intuitive UI that showcases player stats and team information clearly
Integrating Claude with MCP to leverage both AI reasoning and real NBA data
Building an engaging simulation that feels reasonably authentic to NBA basketball within a very short timeframe

What I learned

How to effectively use Model Context Protocol to extend Claude's capabilities with external tools
How to have many AI agents communicate among themselves and with the user