The #1 Open Source AI Data Catalog

Enterprise-grade metadata platform enabling discovery, governance, and observability across your entire data ecosystem

🤖 NEW: Connect AI Agents to DataHub via Model Context Protocol (MCP)

Connect your AI coding assistants (Cursor, Claude Desktop, Cline) directly to DataHub. Query metadata with natural language: "What datasets contain PII?" or "Show me lineage for this table"

Quick setup:

npx -y @acryldata/mcp-server-datahub init

Learn more →

What is DataHub?

🔍 Finding the right DataHub? This is the open-source metadata platform at datahub.com (GitHub: datahub-project/datahub). It was previously hosted at datahubproject.io, which now redirects to datahub.com. This project is not related to datahub.io, which is a separate public dataset hosting service. See the FAQ below.

DataHub is the #1 open-source AI data catalog that enables discovery, governance, and observability across your entire data ecosystem. Originally built at LinkedIn, DataHub now powers data discovery at thousands of organizations worldwide, managing millions of data assets.

The Challenge: Modern data stacks are fragmented across dozens of tools—warehouses, lakes, BI platforms, ML systems, AI agents, orchestration engines. Finding the right data, understanding its lineage, and ensuring governance is like searching through a maze blindfolded.

The DataHub Solution: DataHub acts as the central nervous system for your data stack—connecting all your tools through real-time streaming or batch ingestion to create a unified metadata graph. Unlike static catalogs, DataHub keeps your metadata fresh and actionable—powering both human teams and AI agents.

Why DataHub?

• 🚀 Battle-Tested at Scale: Born at LinkedIn to handle hyperscale data, now proven at thousands of organizations worldwide managing millions of data assets
• ⚡ Real-Time Streaming: Metadata updates in seconds, not hours or days
• 🤖 AI-Ready: Native support for AI agents via MCP, LLM integrations, and context management
• 🔌 Pioneering Ingestion Architecture: Flexible push/pull framework (widely adopted by other catalogs) with 80+ production-grade connectors extracting deep metadata—column lineage, usage stats, profiling, and quality metrics
• 👨‍💻 Developer-First: Rich APIs (GraphQL, OpenAPI), Python + Java SDKs, CLI tools
• 🏢 Enterprise Ready: Battle-tested security, authentication, authorization, and audit trails
• 🌍 Open Source: Apache 2.0 licensed, vendor-neutral, community-driven

🧠 The Context Foundation

Essential for modern data teams and reliable AI agents:

• Context Management Is the Missing Piece in the Agentic AI Puzzle - Why context management is essential for deploying reliable AI agents at scale
• Data Lineage: What It Is and Why It Matters - Understanding the map of how data flows through your organization
• What is Metadata Management? - A comprehensive guide for enterprise data leaders

📑 Table of Contents

• FAQ
• See DataHub in Action
• Quick Start
• Installation Options
• Architecture
• Use Cases & Examples
• Trusted By
• Ecosystem
• Community
• Contributing
• Resources
• License

❓ Frequently Asked Questions

No. datahub.io is a completely separate project — a public dataset hosting service with no affiliation to this project. DataHub (this project) is an open-source metadata platform for data discovery, governance, and observability, hosted at datahub.com and developed at github.com/datahub-project/datahub.

DataHub was previously hosted at datahubproject.io. That domain now redirects to datahub.com. All documentation has moved to docs.datahub.com. If you find references to datahubproject.io in blog posts or tutorials, they refer to this same project — just under its former domain.

Yes. DataHub was originally built at LinkedIn to manage metadata at scale across their data ecosystem. LinkedIn open-sourced DataHub in 2020. It has since grown into an independent community project under the datahub-project GitHub organization, now hosted at datahub.com.

pip install acryl-datahub
datahub docker quickstart

See the Quick Start section below for full instructions. The PyPI package is acryl-datahub.

🎨 See DataHub in Action

▶️ Watch DataHub in Action:

• 5-Minute Product Tour (YouTube)
• Try Live Demo (No installation required)

🚀 Quick Start

Option 1: Try the Hosted Demo (Fastest)

No installation required. Explore a fully-loaded DataHub instance with sample data instantly:

🌐 Launch Live Demo: demo.datahub.com

Option 2: Run Locally with Python (Recommended)

Get DataHub running on your machine in under 2 minutes:

# Prerequisites: Docker Desktop with 8GB+ RAM allocated

# Upgrade pip and install DataHub CLI
python3 -m pip install --upgrade pip wheel setuptools
python3 -m pip install --upgrade acryl-datahub

# Launch DataHub locally via Docker
datahub docker quickstart

# Access DataHub at http://localhost:9002
# Default credentials: datahub / datahub

Note: You can also use uv or other Python package managers instead of pip.

What's included:

• ✅ Full Stack: GMS backend, React UI, Elasticsearch, MySQL, and Kafka.
• ✅ Sample Data: Pre-loaded datasets, lineage, and owners for exploration.
• ✅ Ingestion Ready: Fully prepared to connect your own local or cloud data sources.

Option 3: Run from Source (For Contributors)

Best for advanced users who want to modify the core codebase or run directly from the repository:

# Clone the repository
git clone https://github.com/datahub-project/datahub.git
cd datahub

# Start all services with docker-compose
./docker/quickstart.sh

# Access DataHub at http://localhost:9002
# Default credentials: datahub / datahub

Next Steps

• 🔌 Connect Your Data: Explore our Ingestion Guides for Snowflake, BigQuery, dbt, and more.
• 📚 Learn the Basics: Walk through the Getting Started Guide
• 🎓 DataHub Academy: Deep dive with our Advanced Tutorials

📦 Installation Options

DataHub supports three deployment models:

• Managed SaaS (DataHub Cloud) — zero infrastructure, SLA-backed, enterprise-ready
• Self-hosted via Docker — ideal for development and small teams
• Kubernetes (Helm) — recommended for production self-hosted deployments

→ See all deployment guides (AWS, Azure, GCP, environment variables)

🏗️ Architecture Overview

• ✅ Streaming-First: Real-time metadata updates via Kafka
• ✅ API-First: All features accessible via APIs
• ✅ Extensible: Plugin architecture for custom entity types
• ✅ Scalable: Proven to 10M+ assets and O(1B) relationships at LinkedIn and other companies in production
• ✅ Cloud-Native: Designed for Kubernetes deployment

→ Full architecture breakdown: components, storage layer, APIs, and design decisions

💻 Use Cases & Examples

Use Case: Extract table metadata, column schemas, and usage statistics from Snowflake data warehouse.

Prerequisites:

• DataHub instance running (local or remote)
• Snowflake account with read permissions
• DataHub CLI installed (pip install 'acryl-datahub[snowflake]')

# snowflake_recipe.yml
source:
  type: snowflake
  config:
    # Connection details
    account_id: "xy12345.us-east-1"
    warehouse: "COMPUTE_WH"
    username: "${SNOWFLAKE_USER}"
    password: "${SNOWFLAKE_PASSWORD}"

    # Optional: Filter specific databases
    database_pattern:
      allow:
        - "ANALYTICS_DB"
        - "MARKETING_DB"

sink:
  type: datahub-rest
  config:
    server: "http://localhost:8080"

# Run ingestion
datahub ingest -c snowflake_recipe.yml

# Expected output:
# ✓ Connecting to Snowflake...
# ✓ Discovered 150 tables in ANALYTICS_DB
# ✓ Discovered 75 tables in MARKETING_DB
# ✓ Ingesting metadata...
# ✓ Successfully ingested 225 datasets to DataHub

What gets ingested:

• Table and view schemas (columns, data types, descriptions)
• Table statistics (row counts, size, last modified)
• Lineage information (upstream/downstream tables)
• Usage statistics (query frequency, top users)

Use Case: Programmatically search DataHub catalog and retrieve dataset metadata.

Prerequisites:

• DataHub instance accessible
• Python 3.8+ installed
• DataHub Python package installed (pip install 'acryl-datahub[datahub-rest]')

from datahub.ingestion.graph.client import DatahubClientConfig, DataHubGraph

# Initialize DataHub client
config = DatahubClientConfig(server="http://localhost:8080")
graph = DataHubGraph(config)

# Search for datasets containing "customer"
# Returns up to 10 most relevant results
results = graph.search(
    entity="dataset",
    query="customer",
    count=10
)

# Process and display results
for result in results:
    print(f"Found: {result.entity.urn}")
    print(f"  Name: {result.entity.name}")
    print(f"  Platform: {result.entity.platform}")
    print(f"  Description: {result.entity.properties.description}")
    print("---")

# Example output:
# Found: urn:li:dataset:(urn:li:dataPlatform:snowflake,analytics.customer_profiles,PROD)
#   Name: customer_profiles
#   Platform: snowflake
#   Description: Aggregated customer data from CRM and transactions
# ---

Response format: Each result contains:

• urn: Unique resource identifier for the dataset
• name: Human-readable dataset name
• platform: Source platform (snowflake, bigquery, etc.)
• properties: Additional metadata (description, tags, owners, etc.)

Use Case: Retrieve upstream and downstream dependencies for a specific dataset.

Prerequisites:

• DataHub GMS endpoint accessible
• Dataset URN available from search or ingestion

GraphQL Query:

query GetLineage {
  dataset(
    urn: "urn:li:dataset:(urn:li:dataPlatform:snowflake,analytics.customer_profiles,PROD)"
  ) {
    # Get upstream dependencies (source tables)
    upstream: lineage(input: { direction: UPSTREAM }) {
      entities {
        urn
        ... on Dataset {
          name
          platform {
            name
          }
        }
      }
    }

    # Get downstream dependencies (consuming tables/dashboards)
    downstream: lineage(input: { direction: DOWNSTREAM }) {
      entities {
        urn
        type
        ... on Dataset {
          name
          platform {
            name
          }
        }
        ... on Dashboard {
          dashboardId
          tool
        }
      }
    }
  }
}

Execute via cURL:

curl -X POST http://localhost:8080/api/graphql \
  -H "Content-Type: application/json" \
  -d '{"query": "query GetLineage { ... }"}'

Response structure:

• upstream: Array of datasets that feed into this dataset
• downstream: Array of datasets, dashboards, or ML models that consume this dataset
• Each entity includes URN, type, and basic metadata

Use Case: Programmatically add or update dataset documentation and custom properties.

Prerequisites:

• DataHub Python SDK installed
• Write permissions to DataHub instance
• Dataset already exists in DataHub (from ingestion)

from datahub.metadata.schema_classes import DatasetPropertiesClass
from datahub.emitter.mce_builder import make_dataset_urn
from datahub.emitter.rest_emitter import DatahubRestEmitter

# Create emitter to send metadata to DataHub
emitter = DatahubRestEmitter("http://localhost:8080")

# Create dataset URN (unique identifier)
dataset_urn = make_dataset_urn(
    platform="snowflake",
    name="analytics.customer_profiles",
    env="PROD"
)

# Define dataset properties
properties = DatasetPropertiesClass(
    description="""
    Customer profiles aggregated from CRM and transaction data.

    **Update Schedule:** Updated nightly via Airflow DAG `customer_profile_etl`
    **Data Retention:** 7 years for compliance
    **Owner:** Data Platform Team
    """,
    customProperties={
        "owner_team": "data-platform",
        "update_frequency": "daily",
        "data_sensitivity": "PII",
        "upstream_dag": "customer_profile_etl",
        "business_domain": "customer_analytics"
    }
)

# Emit metadata to DataHub
emitter.emit_mcp(
    entityUrn=dataset_urn,
    aspectName="datasetProperties",
    aspect=properties
)

print(f"✓ Successfully updated documentation for {dataset_urn}")

What this does:

1. Adds rich markdown documentation visible in DataHub UI
2. Sets custom properties for governance and discovery
3. Makes dataset searchable by custom property values
4. Enables filtered searches (e.g., "show me all PII datasets")

Use Case: Enable AI agents (Cursor, Claude Desktop, Cline) to query DataHub metadata directly from your IDE or development environment.

Prerequisites:

• DataHub instance running and accessible
• MCP-compatible AI tool installed (Cursor, Claude Desktop, Cline, etc.)
• Node.js 18+ installed

Quick Setup:

# Initialize MCP server for DataHub
npx -y @acryldata/mcp-server-datahub init

# Follow the interactive prompts to configure:
# - DataHub GMS endpoint (e.g., http://localhost:8080)
# - Authentication token (if required)
# - MCP server settings

Configure your AI tool:

For Claude Desktop, add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "datahub": {
      "command": "npx",
      "args": ["-y", "@acryldata/mcp-server-datahub"]
    }
  }
}

For Cursor, configure in Settings → Features → MCP Servers

What you can ask your AI:

• "What datasets contain customer PII in production?"
• "Show me the lineage for analytics.revenue_table"
• "Who owns the 'Revenue Dashboard' in Looker?"
• "Find all datasets in the marketing domain"
• "What's the schema for user_events table?"
• "List datasets tagged as 'critical' or 'sensitive'"

Example conversation:

You: "What datasets are owned by the data-platform team?"

AI: Based on DataHub metadata, here are the datasets owned by data-platform:
- urn:li:dataset:(urn:li:dataPlatform:snowflake,analytics.customer_profiles,PROD)
  Name: customer_profiles
  Platform: Snowflake
  Description: Aggregated customer data from CRM and transactions

- urn:li:dataset:(urn:li:dataPlatform:bigquery,marketing.campaign_performance,PROD)
  Name: campaign_performance
  Platform: BigQuery
  Description: Marketing campaign metrics and ROI tracking

[... more results]

Benefits:

• ✅ Query metadata without leaving your IDE
• ✅ Natural language interface (no SQL/GraphQL needed)
• ✅ Real-time access to DataHub's metadata graph
• ✅ Understand data context while coding
• ✅ Discover relevant datasets for your task

📖 Full Documentation: MCP Server for DataHub

Common Use Cases

📝 DataHub in Action

Learn from teams using DataHub in production and get practical guidance:

🏢 Trusted by Industry Leaders

3,000+ organizations run DataHub in production worldwide — across both open-source deployments and DataHub Cloud — from hyperscale tech companies to regulated financial institutions and healthcare providers.

By Industry

🛒 E-Commerce & Retail: Etsy • Experius • Klarna • LinkedIn • MediaMarkt Saturn • Uphold • Wealthsimple • Wolt

🏥 Healthcare & Life Sciences: CVS Health • IOMED • Optum

✈️ Travel & Transportation: Cabify • DFDS • Expedia Group • Hurb • Peloton • Viasat

📚 Education & EdTech: ClassDojo • Coursera • Udemy

💰 Financial Services: Banksalad • Block • Chime • FIS • Funding Circle • GEICO • Inter&Co • N26 • Santander • Shanghai HuaRui Bank • Stash • Visa

🎮 Gaming, Entertainment & Streaming: Netflix • Razer • Showroomprive • TypeForm • UKEN Games • Zynga

🚀 Technology & SaaS: Adevinta • Apple • Digital Turbine • DPG Media • Foursquare • Geotab • HashiCorp • hipages • inovex • KPN • Miro • MYOB • Notion • Okta • Rippling • Saxo Bank • Slack • ThoughtWorks • Twilio • Wikimedia • WP Engine

📊 Data & Analytics: ABLY • DefinedCrowd • Grofers • Haibo Technology • Moloco • PITS Global Data Recovery Services • SpotHero

And thousands more across DataHub Core and DataHub Cloud.

Featured Case Studies

• 📰 Optum: Data Mesh via DataHub
• 🏦 Saxo Bank: Enabling Data Discovery in Data Mesh
• 🚗 SpotHero: Data Discoverability at Scale

Using DataHub? Please feel free to add your organization to the list if we missed it — open a PR or let us know on Slack.

🌐 DataHub Ecosystem

DataHub is part of a rich ecosystem of tools and integrations.

Official Repositories

Community Plugins & Integrations

Integrations by Category

📊 BI & Analytics: Tableau • Looker • Power BI • Superset • Metabase • Mode • Redash

🗄️ Data Warehouses: Snowflake • BigQuery • Redshift • Databricks • Synapse • ClickHouse

🔄 Data Orchestration: Airflow • dbt • Dagster • Prefect • Luigi

🤖 ML Platforms: SageMaker • MLflow • Feast • Kubeflow • Weights & Biases

🔗 Data Integration: Fivetran • Airbyte • Stitch • Matillion

View all 80+ integrations →

💬 Community & Support

Join thousands of data practitioners building with DataHub!

🗓️ Town Halls

Next Town Hall:

• 🎟️ Register for the next Town Hall

Last Town Hall:

• 📺 Powering AI Agents with DataHub Context (January 2026)

→ View all past recordings

💬 Get Help & Connect

📧 Stay Updated

• 📝 Read the Blog - Deep dives and case studies
• 📖 Monthly Release Notes - What's new

🎓 Learning Resources

• DataHub Quickstart - Get started in 15 minutes
• API Documentation - GraphQL & REST API reference
• Architecture Guide - Deep dive into internals
• Video Tutorials - Step-by-step guides

🤝 Contributing

We ❤️ contributions from the community! See CONTRIBUTING.md for setup, guidelines, and ways to get involved.

Browse Good First Issues to get started!

📚 Resources & Learning

📰 Featured Content

Blog Posts & Articles:

• DataHub: Popular Metadata Architectures Explained - LinkedIn Engineering
• Open Sourcing DataHub - LinkedIn Engineering
• Enabling Data Discovery in Data Mesh - Saxo Bank
• Data Discoverability at SpotHero - SpotHero
• Emerging Architectures for Modern Data Infrastructure - a16z

Conference Talks:

• The Evolution of Metadata: LinkedIn's Journey - Strata 2019
• Driving DataOps Culture with DataHub - DataOps Unleashed 2021
• Journey of Metadata at LinkedIn - Crunch Conference 2019
• DataHub Journey with Expedia Group - Expedia

Podcasts:

• Bringing The Power Of The Real-Time Metadata Graph To Everyone - Data Engineering Podcast

🔗 Important Links

📄 License

DataHub is open source software released under the Apache License 2.0.

Copyright 2015-2025 LinkedIn Corporation
Copyright 2025-Present DataHub Project Contributors

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.

What this means:

• ✅ Commercial use allowed
• ✅ Modification allowed
• ✅ Distribution allowed
• ✅ Patent use allowed
• ✅ Private use allowed

Learn more: Choose a License - Apache 2.0