Microsoft Fabric Launches Rayfin BaaS & Databricks Counters with Lakebase & MCP Governance

🔄 Update — July 05, 2026: Developer Tooling and CLI Scaffolding for Project Rayfin

Recent signals from the developer community show accelerating adoption and growing interest in Microsoft Fabric Rayfin. In particular, step-by-step guides—such as Rajeev Pentyala’s tutorial on building apps using the Rayfin CLI—demonstrate how straightforward project scaffolding and deployment have become. This significantly speeds up the provisioning of code-first backend solutions on Fabric.

What’s new?

Rayfin CLI Scaffolding: Developers can now initialize structured projects directly via the Rayfin CLI, generating pre-configured templates with minimal setup.
Detailed Onboarding Guides: New community publications offer step-by-step instructions for getting started with the code-first BaaS model on Microsoft Fabric.

Why this adds to the article

The new CLI scaffolding capabilities and onboarding resources prove that Project Rayfin is maturing past a theoretical framework into active, developer-ready tooling. This reinforces the article’s core thesis that OLAP vendors are successfully lowering the barrier to entry for operational app runtimes, keeping developers within their unified data ecosystems.

Summary

A major paradigm shift is sweeping through the cloud data landscape: leading analytical and data warehousing platforms (OLAP) are aggressively expanding into the operational and transactional app-development market (OLTP). Microsoft Fabric recently introduced “Rayfin”—an open-source, code-first Backend-as-a-Service (BaaS) SDK and CLI tool that turns Fabric into a full application runtime. Simultaneously, Databricks has countered with “Lakebase” (a serverless, managed PostgreSQL engine designed for low-latency operational workloads) and has integrated native Model Context Protocol (MCP) support via Unity Catalog to ensure secure AI agent tool governance. These advancements effectively remove the traditional barriers between transactional apps and analytical estates.

What happened?

Microsoft Fabric Rayfin Launch: Microsoft launched Rayfin, a code-first SDK. Using TypeScript decorators, developers can define data models, auth rules, and security policies, deploying the entire backend (SQL database, Entra ID authentication, APIs via Data API Builder) straight to Fabric using a single command (npx rayfin up). All operational data flows into OneLake without requiring ETL.
Databricks Lakebase Postgres: Databricks introduced Lakebase Postgres—a serverless, highly-available PostgreSQL database featuring decoupled storage and compute. It offers Git-like database branching, point-in-time recovery, and scales down to zero when idle.
Native MCP Governance in Databricks: Databricks implemented the open-source Model Context Protocol (MCP) to standardize how AI agents connect to data and tools. Managed by the Unity AI Gateway, it enforces access permissions defined in Unity Catalog. It supports Managed MCP (SQL, AI Search), External MCP (using OAuth), and Custom MCP (built as Databricks Apps).

Why it matters

Traditionally, application databases (OLTP) and analytical warehouses (OLAP) lived in separate worlds. Developers had to deploy independent relational databases (e.g., RDS or Supabase) and manage complex ETL pipelines to mirror transactional data for analytical or machine learning workloads.

Rayfin and Lakebase eliminate these hurdles:

No More ETL: Transactional application data is saved directly within the governed analytical estate (Fabric’s OneLake or Databricks’ Unity Catalog).
Built-in Governance: Applications automatically inherit enterprise-grade compliance, security, and access control policies from the parent environment.
AI Agent Readiness: AI agents need standard, secure mechanisms to interact with enterprise data. Databricks’ MCP integration serves as a universal connector (acting like a “USB-C port for AI”).

Evidence

Official Documentation: Databricks’ AWS documentation provides step-by-step instructions on setting up and governing MCP servers for AI agents.
Developer Discussions: Technical threads on YouTube and LinkedIn discuss the paradigm shift of turning Microsoft Fabric into an application backend via Rayfin.
Product Demonstrations: Tutorials like “Stop Using Postgres the Old Way” showcase Databricks Lakebase setup, serverless autoscaling, and Git-like database branching.

Analysis

This expansion by OLAP giants alters the dynamics of the cloud development market. Microsoft and Databricks are directly targeting established BaaS providers (like Supabase or Firebase) and traditional cloud databases (RDS). By merging the transactional and analytical layers, they provide a highly integrated single-tenant ecosystem. This is a game-changer for AI agents: a chatbot can persist session states and chat history in Lakebase while safely accessing corporate knowledge using the exact same Unity Catalog security boundary. The primary engineering bottleneck shifts from data integration pipelines to the latency and cost overhead of real-time model requests moving through complex policy layers.

Practical Takeaways

For CTOs and software architects, we recommend the following:

Evaluate Rayfin for Internal Tools: If your organization is already on Microsoft Fabric, use Rayfin (npx rayfin up) to quickly deploy secure internal web applications without incurring extra infrastructure management costs.
Utilize Lakebase for Agent Session State: For stateful AI agents within Databricks, Lakebase offers a serverless PostgreSQL option that integrates seamlessly with Unity Catalog.
Standardize Agent Tools with MCP: Implement the Model Context Protocol in your AI agent architectures to ensure tool reusability and centralize security governance through the Unity AI Gateway.

Open Questions

What are the precise latency and performance overheads of real-time model requests when evaluated against multiple policy and permission layers in Unity Catalog?
How will the pricing and storage scaling models for highly write-intensive transactional workloads compare with dedicated OLTP databases?