Skip to content

Demo Walkthrough

Use this walkthrough when you need to show the repo in a workshop or short live demo.

Current demo flow

The goal is to explain the sample through one simple loop:

Choose a deployment mode
  -> Run a retrieve query
    -> Inspect activity and references
      -> Decide whether the evidence proves MCP, Fabric, or combined grounding

Before You Start

Run the local validation gate:

bash scripts/validate-local.sh

For a live deployment, choose one mode:

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

Do not use private tenant values, generated deployment reports, or local screenshots in public docs unless they have been sanitized.

Three-Minute Demo

1. Start With The Mode Selector

Open Live Knowledge Sources Manual and show the three paths:

Mode Message
mcp-only Fastest first run. Validates Microsoft Learn MCP Server KS without Fabric.
byo-fabric Primary Fabric live path. Connects an existing Fabric workspace and ontology.
full Greenfield platform story. Creates Fabric sample assets, then connects Azure AI Search.

Talk track:

This repo packages two public preview live Knowledge Sources into three reusable demo paths.
MCP-only is the quickest validation path.
BYO Fabric is for customers with existing semantic assets.
Full mode is the zero-to-demo platform path when quota and tenant settings are ready.

2. Show The App Overview

Open the deployed app URL from:

deployments/<env>/deployment-summary.md

In the app, start with Current Demo Flow and Deployment Readiness.

What to point out:

  • deployment mode,
  • app runtime status,
  • MCP live path,
  • Fabric live or offline replay status,
  • source trace sections.

Do not show secrets, app settings, raw tokens, or private service URLs in screenshots.

3. Run MCP Live

Use the MCP panel or /api/retrieve/mcp.

Suggested query:

What must be configured to create an Azure AI Search MCP Server knowledge source?

Expected evidence:

  • MCP Server Knowledge Source is microsoft-learn-mcp-ks.
  • Activity or references show Microsoft Learn MCP evidence.
  • The answer is useful, but the proof is the trace: activity, references, and source data.

Claim you can make:

Azure AI Search can call a remote HTTPS MCP server at retrieve time and return traceable evidence.

4. Show Fabric Ontology

Use the Fabric panel or /api/retrieve/fabric.

Suggested query:

Which carriers have the highest customer-care exposure this month?

Expected evidence in live mode:

  • Fabric Ontology Knowledge Source exists.
  • The combined or Fabric KB includes the Fabric source.
  • Live retrieve uses delegated source authorization.
  • Activity or source data shows Fabric Ontology evidence.

Expected behavior without live Fabric auth:

  • The app shows offline replay.
  • The response explains that Fabric live mode needs Fabric IDs and delegated source authorization.

Claim you can make only with live evidence:

Fabric Ontology KS can ground retrieval on governed Fabric semantics with user source authorization.

Safe fallback claim:

Offline replay shows the expected Fabric trace shape while live Fabric auth is being configured.

5. Show Combined Trace

Use the combined panel or /api/retrieve/combined.

Suggested query:

Identify the top customer-care exposure carrier and cite implementation guidance.

What to inspect:

  • source badges or trace sections,
  • activity entries,
  • references,
  • source-specific data,
  • whether the answer used MCP, Fabric, or offline replay.

Claim you can make:

The same Knowledge Base pattern can route across live tools and semantic business data, then expose trace evidence for review.

Ten-Minute Workshop Path

Use this order when the audience needs to reproduce the sample:

  1. Run bash scripts/validate-local.sh.
  2. Read Choose a Pattern.
  3. Deploy mcp-only.
  4. Open the app and run the MCP query.
  5. Open notebooks/01-mcp-server-ks-quickstart.ipynb.
  6. Inspect samples/rest/01-create-mcp-server-ks.http through samples/rest/03-retrieve-mcp.http.
  7. Move to byo-fabric when Fabric workspace and ontology IDs are ready.
  8. Open notebooks/02-fabric-ontology-ks-airline-ops.ipynb.
  9. Run the combined route and inspect source evidence.

Demo Safety Checks

Before showing the demo outside the working team:

[ ] Local validation passes
[ ] The app URL is from a current deployment summary
[ ] Screenshots do not expose tenant IDs, tokens, keys, or private endpoints
[ ] Fabric live claims are backed by Fabric activity or source data
[ ] Offline replay is clearly described as offline replay
[ ] Cleanup evidence exists when deployment behavior is being claimed
[ ] Preview caveats match docs/13-public-preview-limitations.md

Failure Fallbacks

Failure Use this fallback
MCP live call fails Show samples/responses/mcp-retrieve.sample.json and the local validation result.
Fabric token expires Show Fabric offline replay and explain delegated source authorization.
Full deployment is too slow Show the generated E2E report summary and deployment mode diagram.
App is unavailable Use the notebooks and REST files to show the same retrieve sequence.
Public preview behavior changes Stop at the Learn manual boundary and update the sample before making claims.

What Not To Claim

Avoid these during demos:

  • offline replay proves live Fabric retrieval,
  • screenshots alone prove source selection,
  • azd up alone proves retrieve behavior,
  • full mode will work in every tenant and region,
  • local stdio MCP servers can be attached directly to Azure AI Search.

Use Public Preview Limitations and Caveats for safe wording.