Intugle MCP

Transform Fragmented Data into Connected Semantic Data Model

Overview

Intugle’s GenAI-powered open-source Python library builds a semantic data model over your existing data systems. At its core, it discovers meaningful links and relationships across data assets — enriching them with profiles, classifications, and business glossaries. With this connected knowledge layer, you can enable semantic search and auto-generate queries to create unified data products, making data integration and exploration faster, more accurate, and far less manual.

Who is this for?

• Data Engineers & Architects often spend weeks manually profiling, classifying, and stitching together fragmented data assets. With Intugle, they can automate this process end-to-end, uncovering meaningful links and relationships to instantly generate a connected semantic layer.
• Data Analysts & Scientists spend endless hours on data readiness and preparation before they can even start the real analysis. Intugle accelerates this by providing contextual intelligence, automatically generating SQL and reusable data products enriched with relationships and business meaning.
• Business Analysts & Decision Makers are slowed down by constant dependence on technical teams for answers. Intugle removes this bottleneck by enabling natural language queries and semantic search, giving them trusted insights on demand.

Features

• Semantic Data Model - Transform raw, fragmented datasets into an intelligent semantic graph that captures entities, relationships, and context — the foundation for connected intelligence.
• Business Glossary & Semantic Search: Auto-generate a business glossary and enable search that understands meaning, not just keywords — making data more accessible across technical and business users.
• Data Products - Instantly generate SQL and reusable data products enriched with context, eliminating manual pipelines and accelerating data-to-insight.
• Conceptual Search - Generate data product plans from natural language queries, bridging the gap between business questions and executable data product definitions. Learn more in the documentation.

Supported Integrations

Streamlit App

The intugle library includes a Streamlit application that provides an interactive web interface for building and visualizing semantic data models.

https://github.com/user-attachments/assets/402c3f3d-baf3-4ece-ba55-4e06437defc5

To use the Streamlit app, install intugle with the streamlit extra:

pip install intugle[streamlit]

You can launch the Streamlit application using the intugle-mcp command or uvx:

intugle-streamlit
# Or using uvx
uvx --from intugle[streamlit] intugle-streamlit

Open the URL provided in your terminal (usually http://localhost:8501) to access the application. For more details, refer to the Streamlit App documentation.

To run the app in a cloud environment like Google Colab, please refer to our Streamlit quickstart notebook.

Documentation

For more detailed information, advanced usage, and tutorials, please refer to our full documentation site.

Community

Join our community to ask questions, share your projects, and connect with other users.

• Join our Discord

Contributing

Contributions are welcome! Please see the CONTRIBUTING.md file for guidelines.

License

This project is licensed under the Apache License, Version 2.0. See the LICENSE file for details. Third-party software notices are available in the NOTICE file.

Getting Started

Installation

For Windows and Linux, you can follow these steps. For macOS, please see the additional steps in the macOS section below.

Before installing, it is recommended to create a virtual environment:

python -m venv .venv
source .venv/bin/activate

Then, install the package:

pip install intugle

macOS

For macOS users, you may need to install the libomp library:

brew install libomp

If you installed Python using the official installer from python.org, you may also need to install SSL certificates by running the following command in your terminal. Please replace 3.XX with your specific Python version. This step is not necessary if you installed Python using Homebrew.

/Applications/Python\ 3.XX/Install\ Certificates.command

Configuration

Before running the project, you need to configure a LLM. This is used for tasks like generating business glossaries and predicting links between tables.

You can configure the LLM by setting the following environment variables:

• LLM_PROVIDER: The LLM provider and model to use (e.g., openai:gpt-3.5-turbo) following LangChain's conventions
• API_KEY: Your API key for the LLM provider. The exact name of the variable may vary from provider to provider.

Here's an example of how to set these variables in your environment:

export LLM_PROVIDER="openai:gpt-3.5-turbo"
export OPENAI_API_KEY="your-openai-api-key"

Quickstart

For a detailed, hands-on introduction to the project, please see our quickstart notebooks:

Domain	Notebook	Open in Colab
Healthcare	quickstart_healthcare.ipynb
Tech Manufacturing	quickstart_tech_manufacturing.ipynb
FMCG	quickstart_fmcg.ipynb
Sports Media	quickstart_sports_media.ipynb
Databricks Unity Catalog [Health Care]	quickstart_healthcare_databricks.ipynb	Databricks Notebook Only
Snowflake Horizon Catalog [ FMCG ]	quickstart_fmcg_snowflake.ipynb	Snowflake Notebook Only
Native Snowflake with Cortex Analyst [ Tech Manufacturing ]	quickstart_native_snowflake.ipynb
Native Databricks with AI/BI Genie [ Tech Manufacturing ]	quickstart_native_databricks.ipynb
Streamlit App	quickstart_streamlit.ipynb
Conceptual Search	quickstart_conceptual_search.ipynb
Composite Relationships Prediction	quickstart_basketball_composite_links.ipynb

These datasets will take you through the following steps:

• Generate Semantic Model → The unified layer that transforms fragmented datasets, creating the foundation for connected intelligence.
  • 1.1 Profile and classify data → Analyze your data sources to understand their structure, data types, and other characteristics.
  • 1.2 Discover links & relationships among data → Reveal meaningful connections (PK & FK), including composite keys, across fragmented tables.
  • 1.3 Generate a business glossary → Create business-friendly terms and use them to query data with context.
  • 1.4 Enable semantic search → Intelligent search that understands meaning, not just keywords—making data more accessible across both technical and business users.
  • 1.5 Visualize semantic model→ Get access to enriched metadata of the semantic layer in the form of YAML files and visualize in the form of graph
• Build Unified Data Products → Simply pick the attributes across your data tables, and let the toolkit auto-generate queries with all the required joins, transformations, and aggregations using the semantic layer. When executed, these queries produce reusable data products.

Usage

The core workflow of the project involves using the SemanticModel to build a semantic layer, and then using the DataProduct to generate data products from that layer.

from intugle import SemanticModel

# Define your datasets
datasets = {
    "allergies": {"path": "path/to/allergies.csv", "type": "csv"},
    "patients": {"path": "path/to/patients.csv", "type": "csv"},
    "claims": {"path": "path/to/claims.csv", "type": "csv"},
    # ... add other datasets
}

# Build the semantic model
sm = SemanticModel(datasets, domain="Healthcare")
sm.build()

# Access the profiling results
print(sm.profiling_df.head())

# Access the discovered links
print(sm.links_df)

For detailed code examples and a complete walkthrough, please see our quickstart notebooks.

Data Product

Once the semantic model is built, you can use the DataProduct class to generate unified data products from the semantic layer.

from intugle import DataProduct

# Define an ETL model
etl = {
  "name": "top_patients_by_claim_count",
  "fields": [
    {
      "id": "patients.first",
      "name": "first_name",
    },
    {
      "id": "patients.last",
      "name": "last_name",
    },
    {
      "id": "claims.id",
      "name": "number_of_claims",
      "category": "measure",
      "measure_func": "count"
    }
  ],
  "filter": {
    "sort_by": [
      {
        "id": "claims.id",
        "alias": "number_of_claims",
        "direction": "desc"
      }
    ],
    "limit": 10
  }
}

# Create a DataProduct and build it
dp = DataProduct()
data_product = dp.build(etl)

# View the data product as a DataFrame
print(data_product.to_df())

Semantic Search

The semantic search feature allows you to search for columns in your datasets using natural language. It is built on top of the Qdrant vector database.

For full setup instructions (including Docker commands and environment variables), please refer to the Semantic Search Documentation.

Usage

Once you have built the semantic model, you can use the search method to perform a semantic search. The search function returns a pandas DataFrame containing the search results, including the column's profiling metrics, category, table name, and table glossary.

from intugle import SemanticModel

# Define your datasets
datasets = {
    "allergies": {"path": "path/to/allergies.csv", "type": "csv"},
    "patients": {"path": "path/to/patients.csv", "type": "csv"},
    "claims": {"path": "path/to/claims.csv", "type": "csv"},
    # ... add other datasets
}

# Build the semantic model
sm = SemanticModel(datasets, domain="Healthcare")
sm.build()
# Perform a semantic search
search_results = sm.search("reason for hospital visit")

# View the search results
print(search_results)

For detailed code examples and a complete walkthrough, please see our quickstart notebooks.

MCP Server

Intugle includes a built-in MCP (Model Context Protocol) server that exposes your semantic layer to AI assistants and LLM-powered clients. Its main purpose is to allow agents to understand your data's structure by using tools like get_tables and get_schema.

Once your semantic model is built, you can start the server with a simple command:

intugle-mcp

This enables AI agents to programmatically interact with your data context. This also enables vibe coding with the library

For detailed instructions on setting up the server and connecting your favorite client, please see our full documentation.