ScoutIQ Homepage

ScoutIQ, AI Football Scouting Agent for the 2026 World Cup

Inspiration

As the 2026 FIFA World Cup approaches, millions of football fans, journalists, commentators, analysts, and coaches search the internet every day for player statistics, tournament history, tactical insights, and squad information.

While fans can usually find basic statistics through search engines, professionals often need much deeper analysis. A commentator preparing for a match may want to compare a player's current form with previous World Cups. A journalist may need to identify emerging talents. A coach or scout may want to find players with a specific playing style or determine who best replaces a key player in a national team.

I realized that despite the abundance of football data available online, there is no simple AI-powered scouting tool that allows users to ask natural-language questions and instantly receive structured, evidence-based scouting reports.

This inspired me to build ScoutIQ, an AI scouting agent that combines Gemini, Google Cloud, and MongoDB MCP tools to transform football questions into actionable scouting intelligence.

What It Does

ScoutIQ is a multi-step AI agent that performs real reasoning instead of simple question answering.

Users can ask questions such as:

"Who plays like Iniesta in the 2026 World Cup?"
"Compare Mbappé's 2026 form to his 2018 peak."
"Who replaced Griezmann as France's creative midfielder?"

The agent then:

Understands the user's intent using Gemini.
Queries MongoDB through MCP tools.
Performs vector similarity search when style-based matching is required.
Retrieves historical and live tournament data.
Generates a structured scouting report.
Assigns a confidence score based on data quality and coverage.

The result is a professional football scouting assistant capable of supporting journalists, commentators, analysts, and coaches.

How I Built It

ScoutIQ was built entirely on Google Cloud and MongoDB technologies.

AI Agent Layer

The core of the system is powered by Gemini 2.5 Flash using Google Cloud Agent Builder (ADK).

The agent is configured with strict instructions to:

Always use tools before making claims.
Ground responses in real data.
Avoid hallucinating player statistics.
Produce structured scouting reports.
Provide confidence scores for every recommendation.

MongoDB MCP Server

To give the agent access to football intelligence, I built a custom MCP server exposing several scouting tools:

search_players
get_player_profile
get_match_timeline
get_team_players

These tools allow Gemini to retrieve player information, compare performances, analyze match timelines, and perform similarity searches directly from MongoDB Atlas.

Data Pipeline

I built a complete football data pipeline that combines multiple sources:

StatsBomb Open Data (2018 and 2022 World Cups)
Kaggle historical World Cup datasets (1998–2014)
Live 2026 tournament data

The pipeline:

Extracts raw football data.
Normalizes player statistics.
Aggregates tournament profiles.
Generates embeddings using Vertex AI.
Stores everything in MongoDB Atlas.

Vector Search

One of the most important features of ScoutIQ is player style matching.

Using Vertex AI text-embedding-004, I generated embeddings for player tournament profiles and stored them in MongoDB Atlas.

MongoDB Atlas Vector Search enables semantic scouting queries such as:

"Find a creative playmaker with exceptional vision and passing ability."

Instead of matching keywords, ScoutIQ finds players with similar playing styles based on their statistical profiles.

Frontend and Deployment

The application was built with:

React
Vite
FastAPI
MongoDB Atlas
Google Cloud Run

The frontend provides a clean scouting interface, real-time reasoning visualization, confidence indicators, and exportable scouting reports.

Challenges I Faced

Deploying the MCP Server

One of the biggest challenges was deploying and configuring the MCP server on Google Cloud Run.

I had to ensure the server remained accessible, exposed the correct tool endpoints, and could communicate reliably with Agent Builder while maintaining low latency.

Agent and MCP Integration

Connecting Gemini to external MCP tools required significant experimentation.

The agent needed to:

Select the correct tool.
Call the tool with appropriate parameters.
Interpret tool responses correctly.
Continue reasoning after receiving data.

Achieving reliable multi-step reasoning took multiple iterations and prompt refinements.

Reducing Hallucinations

Football scouting requires accuracy.

A major challenge was ensuring the agent never invented statistics or player information.

To address this, I designed strict instructions that force ScoutIQ to use MongoDB tools before generating conclusions and to assign confidence levels based on the underlying data quality.

Building Effective Vector Search

Creating meaningful similarity search required careful work on:

Player profile generation
Embedding strategy
Data normalization
Vector indexing

The goal was to make player comparisons reflect actual playing style rather than simple statistical similarity.

What I Learned

This project gave me hands-on experience with several technologies that were new to me.

I learned how to:

Build AI agents using Google Cloud Agent Builder.
Integrate Gemini with external MCP tools.
Deploy MCP servers on Google Cloud Run.
Design multi-step reasoning workflows.
Generate embeddings using Vertex AI.
Implement semantic search with MongoDB Atlas Vector Search.
Structure football data for retrieval and analysis.
Build reliable AI systems that prioritize grounding and factual accuracy.

Most importantly, I learned how powerful MCP can be when combined with Gemini. Instead of acting like a chatbot, ScoutIQ behaves like a true agent that can reason, retrieve information, and generate actionable insights.