Azure AI Search

The articles goes over creating an assistant on Ejento with the Azure MCP (Model Context Protocol) server to interact with Azure AI Search through natural language. This guide focuses on configuring Azure AI Search to query search indexes and retrieve knowledge using conversational prompts with an Ejento Assistant.

Prerequisites
Azure subscription with AI Search resource
MCP Server deployed and accessible to the assistant
Entra ID app registration (service principal) added to the AI Search with relevant role for permissions
For complete Azure MCP setup and configuration, visit: Azure MCP Server Configuration Guide

Azure AI Search provides enterprise search capabilities for document search, knowledge mining, and semantic search across your data. The tool supports querying indexes, retrieving schema information, and searching knowledge sources.

Tool/Operation	Description
Index: Get index details	Retrieve schema, fields, analyzers, scoring profiles, and properties
Index: Query index	Run search queries against an index using search terms
Knowledge: Get knowledge base	Get details of knowledge bases (retrieval/reasoning over sources/indexes)
Knowledge: Retrieve from KB	Search/query knowledge bases with queries or conversational messages
Knowledge: Get source	Get details of knowledge sources (indexes or external data)
Service: List services	List all AI Search services in your subscription

Tool parameters are defined for each operation. Some are specific to individual tools, while others are global and shared. See Tool parameters documentation for details.

For detailed tool documentation: Azure AI Search Tool Reference

Use Case

Azure AI Search tools, managed via the Azure MCP Server, enable search and retrieval capabilities across your documents. These tools connect natural language queries to indexed data, making it easy to surface relevant information from support articles, internal documentation, manuals, and more.
At the heart of Azure MCP is the Query index tool. This tool is the primary mechanism for Semantic Retrieval Augmented Generation, allowing the Ejento assistant to semantically fetch and present the most relevant content in response to user queries. Whether you’re building a support chatbot, an internal knowledge assistant, or an automated documentation search, the Query index tool provides accurate retrieval of indexed data.

How It Works

User submits a natural language query (e.g., “Search for 'machine learning' in the 'documents' index of my 'my-search-service' service.”).

The Ejento assistant uses the Query index tool to search the appropriate Azure AI Search index.

Relevant documents, articles, or snippets are retrieved and presented to the user.

The assistant can summarize, cite, or further refine results as needed.

Recommended Prompts

Assistant Role

You are a data retrieval assistant specializing in Azure Search indexes. Your primary responsibilities include managing and optimizing search indexes, querying data efficiently, and providing relevant insights from the indexed data. When users request information, you should access the appropriate Azure Search indexes to deliver accurate and timely results. Always ensure clarity in your responses and maintain a focus on user needs. Provide results in a structured format that highlights key data points and insights. If you encounter unclear requests, seek clarification on the specific data or insights the user requires.

Custom Instructions

For azure search related, always assume "parameters": { "service": "Your search service name"}", "index": "Your index name", "query": "{}" } }

Never answer from your own knowledge, ALWAYS CALL the azure mcp tool for searching

Select an appropriate assistant name and description that reflect its purpose. Set the Assistant type to "QnA" for question-and-answer scenarios, and choose a language model that best fits your requirements.

For reasoning pattern, consider the complexity of your use case:

ReAct is recommended for straightforward queries and quick responses.

Reflection provides critique on the responses

DeepAgent provides more advanced planning and reasoning, suitable for complex tasks, but may result in longer response times.

1. Click on Create Assistant

2. Click on Select project...

3. Click your project or create new if you haven't

4. Type "You are an assistant that utilizes Azure MCP - particularly the search service"

5. Click on Add Assistant

7. Click on assistant tools

8. Turn off Rag Tool

9. Turn off Attachment Tool

10. Toggle on Azure MCP

11. Click on Azure MCP

12. Turn off every tool except search

About the Search Tool:
Search operations provide commands for Azure AI Search (formerly "Azure Cognitive Search") services, indexes, and knowledge sources. Use this tool for enterprise search, document search, and knowledge mining. This is a hierarchical MCP command router where sub-commands require specific fields in the "parameters" object. Set "learn=true" to discover available operations for search services and indexes.

13. Click on customize

14. Click on Add custom instruction

15. Add an instruction so the agent knows the index to query

Add this custom instruction with your specific service and index details:

For azure search related, always assume "parameters": { "service": "Your search service name"}", "index": "Your index name", "query": "{}" } }

16. Click on Save

17. Write any natural language question about contents in your index

Best Practices for Ejento AI Integration

Required Parameters

Always provide these parameters in your custom instructions for optimal results:

service: Your Azure AI Search service name

index: The specific search index to query

query: The search query (can be empty {} for the assistant to populate)

Effective Prompting Tips

Be specific: Clearly state what you're searching for

Provide context: Mention the type of documents or data being searched

Use natural language: The assistant translates conversational queries into search syntax

Iterate: Refine queries based on initial results

Example Queries

"Find all documents about machine learning in the products index"

"Search for customer feedback containing 'pricing' or 'cost'"

"Show me the most recent entries in the knowledge base"

Pro Tip: Follow this official documentation to know how to format your query

Use Case#

How It Works#

Recommended Prompts#

Assistant Role#

Custom Instructions#

1. Click on Create Assistant#

2. Click on Select project...#

3. Click your project or create new if you haven't#

4. Type "You are an assistant that utilizes Azure MCP - particularly the search service"#

5. Click on Add Assistant#

6. Click on dropdown trigger#

7. Click on assistant tools#

8. Turn off Rag Tool#

9. Turn off Attachment Tool#

10. Toggle on Azure MCP#

11. Click on Azure MCP#

12. Turn off every tool except search#

13. Click on customize#

14. Click on Add custom instruction#

15. Add an instruction so the agent knows the index to query#

16. Click on Save#

17. Write any natural language question about contents in your index#

Best Practices for Ejento AI Integration#

Required Parameters#

Effective Prompting Tips#

Example Queries#

Use Case

How It Works

Recommended Prompts

Assistant Role

Custom Instructions

1. Click on Create Assistant

2. Click on Select project...

3. Click your project or create new if you haven't

4. Type "You are an assistant that utilizes Azure MCP - particularly the search service"

5. Click on Add Assistant

6. Click on dropdown trigger

7. Click on assistant tools

8. Turn off Rag Tool

9. Turn off Attachment Tool

10. Toggle on Azure MCP

11. Click on Azure MCP

12. Turn off every tool except search

13. Click on customize

14. Click on Add custom instruction

15. Add an instruction so the agent knows the index to query

16. Click on Save

17. Write any natural language question about contents in your index

Best Practices for Ejento AI Integration

Required Parameters

Effective Prompting Tips

Example Queries