Fabric Ontology MCP Server

Production-ready MCP server for full CRUD control of Ontology items in Microsoft Fabric.

Prerequisites

• Python ≥ 3.11
• Azure CLI (az) installed and logged in (az login)
• Access to a Microsoft Fabric workspace with Ontology items

Installation

cd "Ontology MCP Server"
pip install -e .

Usage

# Run via entry point
fabric-ontology-mcp

# Or directly
python -m src

VS Code / Copilot MCP config

{
  "servers": {
    "fabric-ontology": {
      "command": "python",
      "args": ["-m", "src"],
      "cwd": "/path/to/Ontology MCP Server"
    }
  }
}

Available Tools

Workspace Discovery

Tool	Description
list_workspaces	List all Fabric workspaces accessible to you
list_workspace_items	List items in a workspace (filter by type: Eventhouse, Lakehouse, etc.)

Ontology CRUD

Tool	Description
list_ontologies	List ontologies in a workspace
get_ontology	Get ontology metadata
create_ontology	Create a new ontology
update_ontology	Update display name / description
delete_ontology	Delete an ontology (soft or hard)
get_ontology_definition	Get full decoded definition (entities, relationships, bindings)
update_ontology_definition_raw	Replace entire definition from JSON

Entity Types

Tool	Description
list_entity_types	List all entity types in an ontology
get_entity_type	Get a single entity type with its bindings, documents, overviews, links
add_entity_type	Create a new entity type with properties
update_entity_type	Rename an entity type or change its display name property / ID parts
remove_entity_type	Delete an entity type (and its relationships)

Properties

Tool	Description
add_property	Add a property to an entity type
update_property	Rename a property or change its value type
remove_property	Remove a property from an entity type

Relationship Types

Tool	Description
list_relationship_types	List relationships in an ontology
get_relationship_type	Get a single relationship with its contextualizations
add_relationship_type	Create a relationship between entity types (validates both exist)
update_relationship_type	Rename a relationship type
remove_relationship_type	Delete a relationship type

Data Bindings

Tool	Description
list_data_bindings	List bindings for an entity type
add_data_binding	Bind an entity type to a Lakehouse or Eventhouse table
remove_data_binding	Remove a data binding

Documents, Overviews & Resource Links

Tool	Description
add_document	Attach a document URL to an entity type
list_documents	List all documents attached to an entity type
remove_document	Remove a document by URL
get_overview	Get the current overview configuration
set_overview	Configure overview widgets for an entity type
get_resource_links	Get the current resource links
set_resource_links	Link Power BI reports to an entity type

Contextualizations

Tool	Description
list_contextualizations	List contextualizations for a relationship type
add_contextualization	Define how a relationship is materialized from data
remove_contextualization	Remove a contextualization

KQL / Eventhouse Discovery

Tool	Description
get_kql_database_details	Get cluster URI, database name, and parent Eventhouse
list_kql_tables	List tables in a KQL database
get_kql_table_schema	Get column names and types for a table

Lakehouse & Workspace Data Discovery

Tool	Description
discover_lakehouse_tables	List all tables in a Lakehouse (via OneLake Table API)
get_lakehouse_table_schema	Get columns with types and Ontology valueType mapping
discover_workspace_data	Full scan — discover all Lakehouses + Eventhouses, all tables and schemas

Ontology planning workflow: Call discover_workspace_data to scan a workspace, then use the returned schemas to design entity types, properties, relationships, and data bindings using the CRUD tools above.

Data Profiling

Tool	Description
preview_kql_table	Preview first N rows from an Eventhouse table
profile_kql_table	Row count, distinct counts, null rates, sample values, min/max per column
preview_lakehouse_table	Preview first N rows from a Lakehouse table (Spark SQL via Livy)
profile_lakehouse_table	Row count, distinct counts, null rates, sample values, min/max per column

Note: First Lakehouse profiling call takes ~30-60s for Spark session startup. Subsequent queries reuse the session and are fast.

Project Structure

src/
├── __init__.py
├── __main__.py          # python -m src entry point
├── auth.py              # Azure CLI token acquisition (per-resource caching)
├── definition_utils.py  # Base64 encode/decode for ontology definition parts
├── fabric_client.py     # Async Fabric REST API client
├── kusto_client.py      # Async Kusto REST query client
└── server.py            # MCP server with all tools

Input Validation

All tools validate inputs before calling the Fabric API:

• Names must match ^[a-zA-Z][a-zA-Z0-9_-]{0,127}$ (entity types, properties, relationships)
• Value types must be one of: String, Boolean, DateTime, Object, BigInt, Double
• JSON parameters return clear error messages on parse failure
• Entity existence is checked when creating relationships (both source and target must exist)

Authentication

Uses Azure CLI tokens. Make sure you're logged in:

az login

Tokens are cached per resource (Fabric API and Kusto clusters are separate audiences) and automatically refreshed when near expiry.

License

MIT