What Can I List?

Databricks Marketplace supports several asset types — from structured datasets to AI models to interactive agents. Each serves a different customer need, and choosing the right combination shapes what your customers can do with your product.

This page covers what's available to list and a critical constraint to plan around: some asset types are only available to Databricks-to-Databricks (D2D) consumers. Customers on non-Databricks or legacy platforms can only receive tabular data.

Asset types at a glance

Asset type	D2D consumers	Open sharing (non-Databricks)	Best for
Tables	✅	✅	Structured data, analytics workloads
Views (dynamic)	✅	✅	Per-customer filtered data from a single dataset
Volumes	✅	❌	Unstructured files, PDFs, images, training data
Notebooks	✅	❌	Guided onboarding, example queries, tutorials
AI Models (MLflow)	✅	❌	Pre-trained models for inference or fine-tuning
MCP Servers	✅	❌	AI agent tools and API integrations
Databricks Apps	✅	❌	Interactive data applications
Genie Agents	✅ (Beta)	❌	Pre-configured AI analysts

Open sharing consumers are limited to tabular data

If you expect any of your consumers to be on non-Databricks platforms, external tools, or legacy Databricks workspaces without Unity Catalog, they can only receive tables and views. Plan your listing strategy accordingly — you may need a separate listing or access model for those consumers.

Tables

Tables are the foundation of most Marketplace listings. Consumers can query shared tables directly using SQL or Python in their Databricks workspace. Tables can be shared with or without history:

• Without history: Consumers see the current snapshot of the data only.
• With history: Consumers can use time travel and process incremental updates via Change Data Feed (CDF) — useful for pipelines that need to track what changed.

See OpenSharing for technical details on creating and sharing tables.

Views (dynamic)

Dynamic views let you serve different data to different customers from a single underlying dataset, using the current_recipient() function to filter rows and mask columns based on recipient identity.

This is how you build a single listing that gives each customer only their own data — without creating separate tables or shares per customer.

See Dynamic Views & Data Filtering for patterns.

Volumes

Volumes let you share unstructured and semi-structured files alongside your structured data. Common use cases:

• AI training data: Raw files, labeled datasets, embeddings
• Reference materials: Data dictionaries, PDFs, documentation
• Code libraries: Private packages consumers can import into their notebooks
• Media: Images, audio, video for specialized domains

Volumes are read-only for recipients. Only available to D2D consumers.

Notebooks

Shared notebooks are the fastest way to reduce time-to-value for new consumers. Include example queries, analytic workflows, and integration patterns written in SQL, Python, Scala, or R.

Up to ten notebooks can be attached to a single listing. The notebook experience is often the first thing a consumer interacts with after receiving access — a well-crafted notebook directly drives adoption.

See Listings — Sample notebooks for best practices.

AI Models (MLflow)

Publish pre-trained ML and AI models that consumers can deploy for inference, fine-tuning, or experimentation without rebuilding from scratch. Models are stored and shared via Unity Catalog's native MLflow model registry.

Good candidates for model listings:

• Domain-specific classification or prediction models
• Embedding models pre-trained on your proprietary data
• Fine-tuned foundation models for industry verticals

Consumers load, evaluate, and serve shared models directly in their own environment.

MCP Servers

MCP (Model Context Protocol) listings expose tools, APIs, and data sources that AI agents can call. They're connection objects that Agent Bricks and other agents install to gain access to your tools during execution.

MCP is the right format when your value is in actions and integrations, not just data — e.g., triggering workflows, calling proprietary APIs, or enriching agent context at runtime.

See MCP Marketplace Validation for the validation process before publishing.

Databricks Apps

Databricks Apps are interactive applications built on Databricks compute that consumers install and run inside their own workspace. They're the right format when your data product is better experienced as a UI or tool than as raw tables.

See Databricks Apps — Getting Started for the full onboarding process.

Genie Agents

Beta

Genie Agent sharing via OpenSharing is currently in Beta.

A Genie Agent listing delivers a pre-configured AI analyst — with your domain knowledge, business metrics, and example questions built in — directly into a customer's workspace. Instead of sharing tables and waiting for customers to learn your schema, you ship a working AI analyst that's ready to answer questions immediately.

When a consumer mounts the share, they get a local copy of the agent pre-loaded with your data and instructions. They own their local copy and can extend it without affecting your original.

See Sharing Genie Agents for the full guide.

Choosing the right combination

Most providers benefit from combining multiple asset types in a single listing:

Goal	Recommended combination
Fast onboarding for SQL users	Tables + Notebooks
AI-ready data product	Tables + Volumes + Genie Agent
Guided analytics experience	Tables + Notebooks + Genie Agent
Actionable AI integration	MCP Server (standalone or with Tables)
Self-serve interactive tool	Databricks App
Serve both Databricks and non-Databricks consumers	Tables (all consumers) + separate D2D listing for richer assets

What's next

• See Listings for how to create high-quality listing pages
• Review OpenSharing for technical details on each asset type
• Read Sharing Genie Agents to understand the Genie Agent sharing workflow