-- Living Mobile --: November 2026

Sunday, November 1, 2026

How CSA addresses MCP Security

The Cloud Security Alliance (CSA) plays a foundational role in establishing the security infrastructure, threat intelligence, and governance frameworks for the Model Context Protocol (MCP). Recognizing MCP as a critical layer in the "agentic AI control plane," the CSA bridges the gap between rapid open-source adoption and enterprise risk management. [1, 2, 3]

The CSA actively addresses MCP security through several core pillars:

1. The MCP Security Resource Center

The CSA launched the dedicated Model Context Protocol Security Project as an industry-first hub designed specifically to secure AI-to-tool connectivity. This hub coordinates community efforts to research threats, establish standard toolkits, and build operational guardrails. [1, 2]

2. Threat Frameworks & Security Standards

The CSA maps out explicit risk mitigation strategies by developing frameworks that track specific vulnerabilities across the agentic landscape: [1]

Risk Frameworks: The project documents core architectural vulnerabilities, notably publishing the Top 10 MCP Server Security Risks and Top 10 MCP Client Security Risks. [1]
The MCP Security Baseline: The CSA produces standardized checklists (commencing with baseline drafts mapped to existing structures like the Cloud Controls Matrix (CCM), Consensus Assessments Initiative Questionnaire (CAIQ), and the AI Organizational Management framework). [1, 2, 3]

3. Community Intelligence & Ecosystem Auditing

To handle vulnerabilities transparently, the CSA manages public databases tracking the security state of the ecosystem: [1]

vulnerability-db: A centralized public repository tracking known vulnerabilities and security advisories specific to MCP integrations.
audit-db: A public record tracking compliance and independent evaluation results of observable MCP components. [1]

4. Zero Trust Advocacy and Governance

In response to structural concerns—such as optional authorization parameters in the native MCP specification—the CSA explicitly mandates that enterprise deployments treat authentication and authorization as strict baseline requirements, rather than operational enhancements. [1]

Furthermore, the CSA integrates these controls into the STAR (Security Trust Assurance and Risk) program. This allows organizations and cloud vendors offering MCP-enabled services to formally evaluate and demonstrate compliance regarding sandboxing, tool supply-chain integrity, and se

Databricks Unity data catalogue

Databricks Unity Catalog is the unified governance layer for data and AI across the Databricks Lakehouse platform. It provides a centralized console to manage access permissions, track data lineage, discover assets, and audit activity across multiple Databricks workspaces and cloud environments. [1, 2, 3, 4, 5]

Unlike traditional workspace-level governance (like the legacy Hive metastore), Unity Catalog operates at the account level, meaning policies and metadata are defined once globally and enforced everywhere. [4, 5, 6]

1. The Hierarchical Object Model

Unity Catalog structures data and AI assets using a robust three-level namespace to organize permissions and simplify data discovery. [1, 6]

Metastore: The top-level container of metadata. It stores security principals (users/groups) and access permissions. It is deployed at the cloud region level and can be linked across multiple workspaces. [3, 5, 7]
Catalog (catalog): The first layer of the object hierarchy, typically used to separate environments (e.g., prod, dev) or business units (e.g., finance, marketing). [8]
Schema (schema): Also known as a database, this is the second layer. It groups related data objects. [8]
Objects (object): The leaf level containing the actual operational assets. This includes structural Tables and Views, unstructured Volumes (files), ML Models, and user-defined Functions. [1, 3, 9]

2. Core Capabilities & Features

Capability	Description
Centralized Identity & Access	Manage users, groups, and service principals centrally at the account level. Permissions flow downward (inheriting from Catalog → Schema → Table).
ANSI SQL Governance	Grant or revoke permissions using standard SQL queries (e.g., `GRANT SELECT ON TABLE...`) or visually via the Catalog Explorer UI.
Automated Data Lineage	Captures column-level data transformations and flow in real time. Helpful for impact analysis before deleting tables and debugging broken dashboards.
Advanced Security Controls	Native row-level filtering (hiding rows depending on who is looking) and column-level masking (hiding sensitive PII data like credit card numbers).
Lakehouse Federation	Query and govern data sitting outside of Databricks (e.g., in Snowflake, MySQL, PostgreSQL, or SAP systems) without moving or duplicating the underlying data.
Delta Sharing	An open protocol integrated natively to securely share live data assets and AI models with external organizations without data duplication.
Comprehensive Auditing	Immutable, queryable log tables that record exactly who did what, when, and from where for security compliance.

3. Managed vs. External Storage

When working with tables and volumes inside Unity Catalog, assets are split into two storage types:

Managed Storage: Databricks fully manages both the lifecycle and the physical file placement in your cloud storage bucket. If you drop a managed table, the underlying files are deleted automatically. [1, 8]
External Storage: You control the directory path where the data is stored using External Locations and Storage Credentials managed by Unity Catalog. If you drop an external table, only the metadata definition is removed; your cloud files remain safe. [1, 7]