Knowledge Layer & MCP Agent
A reusable enterprise knowledge layer exposed through MCP, so Copilot, custom apps, and future Fabric agents all consume the same grounded KPI, schema, and product context through one governed tool surface.
What was broken
Enterprise knowledge sat scattered across Excel files, decks, KPI dictionaries, schema references, semantic relationships, and product-specific documentation. Different consumers — Copilot Studio, custom apps, future Fabric agents — needed the same trusted business context, but there was no unified, governed, reusable access layer. Every team rebuilt its own RAG flow, producing duplicated work, inconsistent answers, and weak grounding.
How I built it
I built a Knowledge Layer Asset organised into three areas: Knowledge Layer Content Creation, Knowledge Layer Store, and an MCP Layer. Curated business knowledge — global definitions, KPI dictionary, table and column dictionary, semantic relationships, product-specific assets like Market Mirror and MMR — is indexed into Azure AI Search across dedicated indexes (Table & Column, KPI, Definition, Semantic, Product Index Registry). On top sits a Remote MCP Server hosted on Azure Function Apps, authenticated through OAuth 2.0 / Microsoft Entra ID inbound and UAMI / Managed Identity outbound to Azure AI Foundry and Azure AI Search. The MCP tool drives a Product Identifier → Schema Identifier → KPI Identifier agent chain that returns grounded context back to the client.
The reasoning
MCP gives the knowledge layer a single, versioned contract. Any client — Copilot Studio, a custom app, a future Fabric data agent — speaks the same protocol and gets the same grounded answer with the same auth, logging, and guardrails. Splitting product, schema, and KPI identification into focused agents keeps each step independently testable and lets the index registry scale to new products without touching consumers.
How the pieces fit
Key components
- 01Remote MCP server hosted on Azure Function Apps with Managed Identity for backend access
- 02Azure AI Search indexing across KPI, schema, definition, semantic, and product registries
- 03Azure AI Foundry agents for product, schema, and KPI identification
- 04Product Index Registry resolves the right product-specific indexes dynamically
- 05Microsoft Entra ID (OAuth 2.0) for inbound auth, UAMI for backend access to Foundry and Search
- 06Copilot Studio integration via MCP tool ask_market_mirror_agent
- 07Future-ready for Fabric data agent consumption through the same MCP contract
Tech stack
In-browser demo
A scripted, in-browser walkthrough of a real run, traced step by step. Press play to watch the agents fire.