Choose a Pattern

Pattern	Use when	Start here
MCP Server KS	You need a low-friction public preview quickstart or tool-backed live retrieval	samples/rest/01-create-mcp-server-ks.http
Fabric Ontology KS	You need governed business semantics from Microsoft Fabric	samples/rest/04-create-fabric-ontology-ks.http
Combined KB	You need to validate multi-source routing and trace behavior	samples/rest/05-create-combined-kb.http

Deployment Modes

Start with mcp-only unless you already have Fabric workspace and ontology IDs ready. If you are choosing between byo-fabric and full, skim the FAQ before deploying. It calls out the most common points of confusion: offline replay, Fabric source authorization, Fabric quota, and MCP endpoint requirements.

Mode	Use when	Success signal
mcp-only	You want the fastest Azure AI Search MCP Server KS validation without Fabric.	MCP retrieve returns activity or references from microsoft_docs_search.
byo-fabric	You already have a Fabric workspace and ontology and want the validated live Fabric path.	Fabric KS is created and live retrieve works with delegated source authorization, or offline replay explains what is missing.
full	You want a greenfield path that creates the Fabric sample stack and connects it to Azure AI Search.	Fabric IDs are generated, KS/KB assets are created, app loads, and cleanup evidence is recorded.

Command shapes:

bash scripts/deploy.sh --mode mcp-only --env-name liveks-mcp --location eastus

bash scripts/deploy.sh \
  --mode byo-fabric \
  --env-file .env.external.local \
  --env-name liveks-byo \
  --location eastus

bash scripts/deploy.sh \
  --mode full \
  --env-name liveks-full \
  --location eastus \
  --fabric-location westus3

Recommended Order

1. Create the MCP Server KS.
2. Create the MCP-only Knowledge Base.
3. Retrieve from MCP and inspect activity, references, and source data.
4. Add Fabric Ontology KS with BYO Fabric IDs or run --mode full to create the sample Fabric assets.
5. Create a combined Knowledge Base and repeat trace validation.

How To Check A Run

The short version:

• local validation proves the repo shape,
• E2E reports prove create-call-load-delete behavior,
• retrieve traces prove source selection,
• screenshots explain the experience but are not enough by themselves.