---

What is this?

IPL Text2SQL MCP is an agentic AI system that lets you query a full IPL cricket database using plain English. Type a question, and a ReAct agent automatically figures out the right SQL, runs it, and returns a clear answer — no SQL knowledge required.

"Who leads the 2026 Orange Cap right now?"
  → Abhishek Sharma (SRH) with 44 avg runs per match

The system is built on the Model Context Protocol (MCP), which means the SQL tools are served as a standalone MCP server that any MCP-compatible client (Claude Desktop, custom agents, etc.) can connect to.

---

Installation

From PyPI (Recommended)

pip install ipl-text2sql-mcp

From Source

git clone https://github.com/yourusername/ipl-text2sql-mcp.git
cd ipl-text2sql-mcp
pip install -e .

Development Installation

git clone https://github.com/yourusername/ipl-text2sql-mcp.git
cd ipl-text2sql-mcp
pip install -e ".[dev]"

---

Quick Start

1. Set up your database:

   python create_ipl_db.py
   

2. Configure your API key:

   # Create .env file
   echo "OPENROUTER_API_KEY=your_openrouter_api_key_here" > .env
   

3. Run the server:

   # Using the command-line tool
   ipl-text2sql-server
   
   # Or using Python directly
   python text2sql_server.py
   

4. Query from another terminal:

   python react_client.py
   

---

Architecture

┌─────────────────────────────────────────────────────┐
│                    User (CLI)                        │
│              "Who won IPL 2025?"                     │
└───────────────────────┬─────────────────────────────┘
                        │
┌───────────────────────▼─────────────────────────────┐
│              ReAct Agent  (react_client.py)          │
│         GPT-4-turbo + LangGraph + System Prompt      │
└───────────────────────┬─────────────────────────────┘
                        │  MCP over HTTP (streamable)
┌───────────────────────▼─────────────────────────────┐
│           MCP Server  (ipl_text2sql_mcp)             │
│          FastMCP · 4 SQL tools · port 8010           │
└───────────────────────┬─────────────────────────────┘
                        │
┌───────────────────────▼─────────────────────────────┐
│        LangChain SQL Toolkit  ←→  SQLite             │
│               ipl_insights.db                        │
│    Team · Player · Season · Match · Batting · Bowling│
└─────────────────────────────────────────────────────┘

---

Database Coverage

19 seasons · 2008–2026 · 47 players · 10 franchises

Table	Rows (approx.)	Description
Team	10	Franchises — CSK, MI, RCB, KKR, SRH, DC, RR, PBKS, GT, LSG
Player	47	Kohli, Rohit, Dhoni, Bumrah, Warner, Gayle, de Villiers and more
Season	19	Year, winner, runner-up, MVP, status (Completed / In Progress)
Match	~1,200	Venue, date, winner, win type/margin, Man of Match
Batting	~16,000	Runs, balls, fours, sixes, strike rate, dismissal type per innings
Bowling	~7,000	Overs, wickets, runs conceded, economy per innings

Schema Relationships

Season ──< Match >── Team
                 └── Player (ManOfMatch)
Match ──< Batting >── Player
      └── Bowling >── Player
Player >── Team (PrimaryTeam)

Notable data points

• 2025: RCB won maiden title, beat PBKS by 6 runs. Orange Cap: Sai Sudharsan (759 runs). Purple Cap: Prasidh Krishna (25 wickets).
• 2026: Season in progress as of April 28 — Punjab Kings lead standings, ~40 matches played.
• Historical: All winners from Rajasthan Royals (2008) through KKR (2024) accurately mapped.

---

MCP Tools

The server exposes four tools over HTTP that the ReAct agent calls autonomously:

Tool	Input	Description
sql_db_list_tables	(none)	List all tables in the database
sql_db_schema	table_names (comma-separated)	Schema + sample rows for given tables
sql_db_query_checker	query (SQL string)	Validate query before execution
sql_db_query	query (SQL SELECT)	Execute query and return results

The agent follows a strict workflow: list → schema → write SQL → check → execute → answer.

---

Quick Start

Prerequisites

• Python 3.10+
• OpenAI API key

1. Clone

git clone https://github.com/sjsoumil/IPL-Text2SQL-MCP.git
cd IPL-Text2SQL-MCP

2. Install dependencies

pip install mcp langchain langchain-openai langchain-community \
            langchain-mcp-adapters langgraph python-dotenv

3. Set up environment

cp .env.example .env

Edit .env and fill in your key:

OPENAI_API_KEY=sk-...
DB_PATH=ipl_insights.db
PORT=8010
MCP_SERVER_URL=http://localhost:8010/mcp

4. Generate the database

python create_ipl_db.py

Output:

Season 2008 done  [Completed]
...
Season 2026 done  [In Progress]
  Team      :     10 rows
  Player    :     47 rows
  Season    :     19 rows
  Match     :  1,167 rows
  Batting   : 16,338 rows
  Bowling   :  7,002 rows

Created ipl_insights.db

5. Start the MCP server

python text2sql_server.py

Starting IPLText2SQLServer
  URL      : http://localhost:8010/mcp
  Database : ipl_insights.db
  Tools    : ['sql_db_list_tables', 'sql_db_schema', 'sql_db_query_checker', 'sql_db_query']

6. Run the interactive client

Open a second terminal:

python react_client.py

======================================================================
IPL Analytics Assistant (2008-2026)  |  type 'exit' to quit
======================================================================
Your question: _

---

Example Questions

# Run scoring
Who are the top 5 run-scorers of all time in the IPL?
Show Virat Kohli's runs season by season.
Who has the highest strike rate among batters with 500+ runs?

# Wickets & bowling
Who took the most wickets in IPL 2025?
What is Jasprit Bumrah's average economy rate across all seasons?

# Teams & titles
Which team has won the most IPL titles?
Which team has the best win percentage in finals?

# Current season (2026)
Who leads the 2026 Orange Cap right now?
How many matches has Punjab Kings won so far in 2026?

# Records
Who has the most Man of the Match awards?
Which player has hit the most sixes in IPL history?
What is the biggest winning margin in IPL history?

---

Project Structure

IPL-Text2SQL-MCP/
├── text2sql_server.py   # FastMCP server — exposes SQL tools over HTTP
├── react_client.py      # Interactive ReAct agent CLI
├── create_ipl_db.py     # Generates ipl_insights.db from scratch
├── .env.example         # Environment variable template
└── README.md

---

How It Works

1. User types a natural-language question in the CLI.
2. ReAct agent (LangGraph + GPT-4.1-mini) reasons step-by-step, deciding which MCP tool to call next.
3. MCP client (langchain-mcp-adapters) forwards each tool call to the FastMCP server over HTTP.
4. MCP server delegates to LangChain's SQLDatabaseToolkit — which inspects the schema, validates the SQL, and executes it against SQLite.
5. Results flow back up the chain and the agent crafts a plain-English answer.

The agent is guided by a detailed system prompt that enforces safe SQL practices (SELECT only), handles ties correctly, and knows the full schema upfront to minimize unnecessary tool calls.

---

Requirements

Dependency	Purpose
mcp	MCP server runtime (FastMCP)
langchain	Core framework
langchain-openai	GPT-4.1-mini LLM
langchain-community	SQL database utilities
langchain-mcp-adapters	MCP ↔ LangChain tool bridge
langgraph	ReAct agent execution
python-dotenv	Environment variable loading