Overview

This accelerator focuses on two live, query-time Knowledge Source patterns for Azure AI Search and Foundry IQ:

• MCP Server Knowledge Source for remote HTTPS MCP tools.
• Fabric Ontology Knowledge Source for governed business semantics in Microsoft Fabric.

Classic retrieval samples often begin by indexing content. This repo shows another pattern: a Knowledge Base can call live sources during retrieval, return answer text, and expose trace evidence through activity, references, and source-specific data.

What This Repo Is

Reusable sample accelerator
  -> deployment modes
  -> REST samples
  -> notebooks
  -> demo app
  -> offline replay
  -> reviewer evidence and safe-claim guidance

The repo is designed for field demos, customer workshops, private product review, blog preparation, and official-sample readiness work. It is not a production reference architecture.

The Two Knowledge Source Patterns

Pattern	What it does	First place to look
MCP Server KS	Calls explicitly allowed tools on a remote HTTPS MCP server at retrieve time.	MCP Server Knowledge Source
Fabric Ontology KS	Grounds retrieval in Fabric ontology entities, relationships, and governed semantic definitions.	Fabric Ontology Knowledge Source
Combined KB	Shows how one Knowledge Base can route across both live source types.	Combined Knowledge Base Routing

Deployment Modes

Start with mcp-only unless you already have Fabric workspace and ontology IDs.

Mode	Purpose	Best for
mcp-only	Deploy Azure AI Search, Azure OpenAI, Microsoft Learn MCP Server KS, MCP-only KB, Search index, and demo app.	First run and low-friction validation.
byo-fabric	Deploy the Azure side and connect an existing Fabric workspace and ontology.	Customer or field demos with existing Fabric semantic assets.
full	Create sample Fabric assets, then deploy Azure AI Search, both Knowledge Source paths, and the demo app.	Greenfield platform story when quota and tenant settings are ready.

For command examples, see Choose a Pattern and One-Command Demo Deployment.

Documentation Map

Use this map when you are reviewing the repo or deciding what to read next.

Need	Read
Understand the architecture	Architecture
Pick the right path	Choose a Pattern
Learn MCP Server KS	MCP Server Knowledge Source
Learn Fabric Ontology KS	Fabric Ontology Knowledge Source
Understand combined routing	Combined Knowledge Base Routing
Review security posture	Security and Governance
Debug a run	Troubleshooting
Pick test questions	Test Queries And Expected Traces
Inspect traces without live resources	Offline Replay
Deploy the app and resources	One-Command Demo Deployment
Connect existing Fabric assets	Fabric Live BYO Validation
Avoid unsafe preview claims	Public Preview Limitations and Caveats
Run a short demo	Demo Walkthrough
Answer common setup questions	FAQ

Validation Loop

Every path in this repo should be reviewed through the same loop:

Create Knowledge Source
  -> attach it to a Knowledge Base
    -> retrieve with a test question
      -> inspect activity, references, and sourceData
        -> record sanitized evidence

The final answer alone is not enough. Good evidence shows which source was selected, what live source was called, and whether cleanup completed when deployment behavior is being claimed.

First-Time Reader Path

1. Read the mode selector in Live Knowledge Sources Manual.
2. Use Choose a Pattern to pick a path, and keep FAQ open for mode, auth, offline replay, and endpoint questions.
3. Run:

bash scripts/validate-local.sh

1. Start with mcp-only.
2. Open the demo app and inspect source trace evidence.
3. Move to byo-fabric when Fabric workspace and ontology IDs are ready.
4. Use full only when Fabric quota, tenant settings, region, and delegated auth expectations are clear.

Evidence And Safety

Generated outputs stay under ignored paths:

.deployment/
deployments/
scratch/

Do not commit tenant IDs, tokens, keys, generated reports, private service URLs, local screenshots, or internal planning notes.

Before sharing claims in a blog, presentation, or customer-facing demo, review:

• Public Preview Limitations and Caveats