⚾ Baseball Dugout

Give your coaching staff an AI-powered scout that researches players, generates reports, and shares them via email—all from a single conversation.

The Use Case

A coach needs to evaluate a player before a trade deadline. Instead of digging through spreadsheets, querying databases, and manually formatting reports:

1. Ask the agent: "Research David Ortiz"
2. The agent: Pulls career stats from Snowflake, creates a scouting report in Google Docs, and emails it to the coaching staff

The same pattern extends to any workflow: analyzing team schedules via Google Calendar, comparing players before draft picks, generating weekly performance summaries—AI agents that integrate with your existing tools.

How It Works

This project demonstrates building custom MCP tools with Arcade to securely connect AI agents to your data and productivity apps:

• Snowflake — Your data warehouse (player stats, team records)
• Arcade — Securely deploys tools + handles auth for Google Docs, Gmail, Calendar
• LangChain — Orchestrates the agent

┌──────────────────┐      ┌──────────────────┐      ┌──────────────────┐
│   LangChain      │ ───▶ │  Arcade MCP      │ ───▶ │   Snowflake      │
│   Agent          │      │  Gateway         │      │   Google Docs    │
└──────────────────┘      └──────────────────┘      │   Gmail          │
                                                    └──────────────────┘

Why Arcade? Your Snowflake credentials stay as secrets—never exposed to the LLM. Google OAuth happens just-in-time when needed. The agent calls tools like GetPlayerStats("Babe Ruth") instead of writing SQL.

Quick Start

1. Set Up Snowflake

Create a free Snowflake account at snowflake.com.

Download the dataset: Baseball Databank from Kaggle.

Run SQL scripts in order (in Snowsight web UI):

snowflake/
├── 01_setup_database.sql   # Create database, warehouse, schema
├── 02_create_tables.sql    # Create all baseball tables
├── 03_create_stage.sql     # Create stage for CSV upload
└── 04_load_data.sql        # Load CSV data into tables

Upload CSV files (between scripts 03 and 04):

1. Go to: Data → Databases → BASEBALL_ANALYTICS → PLAYER_DATA → Stages → BASEBALL_DATA_STAGE
2. Click "+ Files" → Upload: Master.csv, Batting.csv, Pitching.csv, Teams.csv, HallOfFame.csv, AllstarFull.csv, Salaries.csv, TeamsFranchises.csv

2. Deploy MCP Tools to Arcade

cd mcp

# Install Arcade CLI
pip install arcade-ai

# Login to Arcade
arcade login

# Set Snowflake credentials as secrets
arcade secrets set SNOWFLAKE_ACCOUNT your-account-id
arcade secrets set SNOWFLAKE_USER your_username
arcade secrets set SNOWFLAKE_PASSWORD your_password

# Deploy the MCP server
arcade deploy -e server.py

This can also be configured using Arcade Web UI

After deployment, create an MCP gateway in Arcade Dashboard and note the URL.

3. Run the Agent

cd agent

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Configure environment
cp ../env.example ../.env
# Edit .env with your credentials

# Run
python baseball_agent.py

The agent will prompt you for:

1. Player name to research
2. Email address to send the report to

It will then create a Google Doc with the scouting report and email it.

MCP Tools

Tool	Description
GetPlayerStats	Career batting/pitching stats for a player
GetTeamStats	Team history and records
ComparePlayers	Head-to-head player comparison
GetSeasonLeaders	League leaders for a stat/year
ExecuteBaseballQuery	Custom SQL queries

Project Structure

baseball-dugout/
├── snowflake/             # SQL setup scripts
│   ├── 01_setup_database.sql
│   ├── 02_create_tables.sql
│   ├── 03_create_stage.sql
│   └── 04_load_data.sql
├── mcp/                   # MCP tools (deploy to Arcade)
│   ├── server.py
│   └── pyproject.toml
├── agent/                 # LangChain agent
│   ├── baseball_agent.py
│   └── requirements.txt
├── env.example
└── README.md

Data

Baseball dataset: Baseball Databank (Public Domain)

License

MIT