Azure AI Search

You run an MCP server that connects Azure AI Search to LangGraph workflows, enabling natural, conversational access to enterprise documents through Claude Desktop. It orchestrates retrieval, formatting, and analysis with Gemini enhancements, while preserving observability and secure configuration through environment variables. Use it to power intelligent document search, summarization, and analysis in an enterprise setting.

How to use

Deploy the MCP server locally or in your environment and connect it to Claude Desktop. You will issue natural-language prompts asking to search Azure AI Search, retrieve document context, summarize results, or analyze findings. The server exposes tools that Claude can call to perform searches, fetch context, and generate structured outputs or concise analyses. Use LangGraph workflows to manage context, routing, and formatting, and enable LangSmith tracing if you want to monitor and debug the chain executions.

Typical usage patterns include: searching for specific documents, retrieving contextual details for a document, and producing summarized or analyzed results. You can also combine search results with structured prompts to drive deeper insights, comparative analyses, or executive summaries.

How to install

Prerequisites: Python 3.8+ and an Azure AI Search service with an index ready for querying. You may also optionally configure Google Gemini for enhanced processing and LangSmith for observability.

Step 1. Clone the project and install the package. Use the repository’s workflow to install in editable mode.

Step 2. Create your environment file by copying the template and filling in credentials.

Step 3. Run the MCP server to start listening for Claude Desktop requests.

Configuration and getting started

Environment variables you use to configure the MCP server include credentials for Azure AI Search, Google Gemini, and LangSmith. Create a file named .env in the project root and populate it with the following variables as shown. You should replace placeholders with your actual keys and identifiers.

# Azure AI Search
AZURE_SEARCH_ENDPOINT=https://your-search-service.search.windows.net
AZURE_SEARCH_API_KEY=your-search-admin-key
AZURE_SEARCH_INDEX_NAME=your-index-name

# Google Gemini (for enhanced AI processing)
GOOGLE_API_KEY=your-google-api-key
GEMINI_MODEL=gemini-1.5-flash
GEMINI_TEMPERATURE=0.1

# LangSmith (for tracing and debugging)
LANGCHAIN_TRACING_V2=true
LANGCHAIN_ENDPOINT=https://api.smith.langchain.com
LANGCHAIN_API_KEY=your-langsmith-api-key
LANGCHAIN_PROJECT=azure-search-mcp

Google Gemini setup

To enable Gemini-based processing, obtain a Gemini API key and add it to your environment. If you provide the key, the server uses Gemini to format, summarize, and analyze documents. If no API key is present, the server still functions with basic formatting.

Steps to configure Gemini: Get an API key from the Gemini provider, then set GOOGLE_API_KEY in your .env file and/or adjust GEMINI_MODEL and GEMINI_TEMPERATURE as needed.

Claude Desktop integration

Claude Desktop communicates with the MCP server through a JSON-based configuration. Add an MCP server entry to Claude Desktop so Claude can invoke the server as a tool.

{
  "mcpServers": {
    "azure_search_mcp": {
      "command": "python",
      "args": [
        "-m",
        "azure_search_mcp"
      ],
      "cwd": "../langgraph-claude-azure-mcp",
      "env": {
        "PYTHONPATH": "../langgraph-claude-azure-mcp/src"
      }
    }
  }
}

Project structure and development notes

The server is organized to separate configuration, chain logic, Azure AI Search integration, and MCP interfaces. You will interact with the server through Claude Desktop as a tool, while development and testing use the provided test suite and visualization utilities.

Testing and monitoring

Run the test suite to validate chain behavior, integration, and environment loading. Enable LangSmith tracing to monitor tool calls, Azure queries, and LangGraph execution traces.

Troubleshooting

Common issues include startup failures, authentication errors with Azure Search, or Claude not recognizing the MCP tool. Ensure your .env file contains valid credentials, restart Claude Desktop after updating configuration, and verify that the Azure Search index has documents to query.

Available tools

search_documents

Search for relevant documents in Azure AI Search using text queries

get_document_context

Retrieve detailed context from specific documents by their IDs

search_and_summarize

Search and get a summarized view of results

search_with_analysis

Search with relevance analysis