Experimental MCP Service for Azure AI Search

@projectAcetylcholine

About Experimental MCP Service for Azure AI Search

An experimental MCP service for Azure AI Search in Python

#python #tools #ai #mcp #templates #resources #agents #prompts #ai-agents #data-sources #ai-search #indexers #azure-ai-search #mcp-server #mcp-tools #search-documents

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "mcp-server-azure-ai-search-python-preview": {
      "command": "uv",
      "args": [
        "build"
      ]
    }
  }
}

Tools

READ_INDEX

WRITE_INDEX

WRITE_DOCUMENTS

READ_DOCUMENTS

READ_INDEXER

WRITE_INDEXER

READ_INDEXER

FETCH_FILE_CONTENTS

Overview

What is Experimental MCP Service for Azure AI Search?

Experimental MCP Service for Azure AI Search is a Model Context Protocol server that allows MCP hosts to interact with an Azure AI Search service. It provides tools for reading and writing indices, managing documents, and configuring indexers, data sources, and skill sets. Designed for developers who need to automate or integrate Azure AI Search operations within MCP‑enabled environments.

How to use Experimental MCP Service for Azure AI Search?

Install Python 3.12+ and uv. Build the package from source, install the wheel, then run the service using either SSE transport (uv run -m mcp_server_azure_ai_search_preview --transport sse --envFile .env) or STDIO transport (configured in your MCP host’s mcp.json). Set required environment variables for authentication (service principal or API key) and the search endpoint. Optionally filter the exposed tools by setting AZURE_AI_SEARCH_MCP_TOOL_GROUPS to a comma‑delimited list of tool groups.

Key features of Experimental MCP Service for Azure AI Search

List and describe indices and their schemas.
Create, modify, and delete indices.
Add, query, and delete documents.
Manage indexers, data sources, and skill sets.
Fetch contents from local files or URLs.
Supports service principal and API key authentication.

Use cases of Experimental MCP Service for Azure AI Search

Automating index creation and schema changes during development.
Running search queries against an Azure AI Search index from an MCP host.
Setting up and maintaining indexers and data sources in a CI/CD pipeline.
Integrating Azure AI Search management into AI‑assisted coding tools (e.g., VS Code, Claude Desktop).

FAQ from Experimental MCP Service for Azure AI Search

What prerequisites are needed to run the service?

You need Python 3.12 or later, and uv must be installed. You also need a service principal or an API key for the Azure AI Search service, along with the appropriate environment variables.

How do I install and run the service?

Build the package with uv build, install the wheel with pip install ..., then run it using one of the provided commands for SSE or STDIO transport. All tools are currently marked experimental and may change without notice.

What authentication methods are supported?

Two methods are supported: service principal authentication (requires AZURE_TENANT_ID, AZURE_CLIENT_ID, AZURE_CLIENT_SECRET) and API key authentication (requires AZURE_AI_SEARCH_API_KEY). Set AZURE_AUTHENTICATION_METHOD to "service-principal" or "api-search-key".

Can I limit which tools are exposed to my MCP host?

Yes. Set the environment variable AZURE_AI_SEARCH_MCP_TOOL_GROUPS to a comma‑separated list of tool groups (e.g., READ_INDEX,READ_DOCUMENTS). Use ALL to include every tool.

Frequently asked questions

What prerequisites are needed to run the service?

You need Python 3.12 or later, and `uv` must be installed. You also need a service principal or an API key for the Azure AI Search service, along with the appropriate environment variables.

How do I install and run the service?

Build the package with `uv build`, install the wheel with `pip install ...`, then run it using one of the provided commands for SSE or STDIO transport. All tools are currently marked experimental and may change without notice.

What authentication methods are supported?

Two methods are supported: **service principal** authentication (requires `AZURE_TENANT_ID`, `AZURE_CLIENT_ID`, `AZURE_CLIENT_SECRET`) and **API key** authentication (requires `AZURE_AI_SEARCH_API_KEY`). Set `AZURE_AUTHENTICATION_METHOD` to `"service-principal"` or `"api-search-key"`.

Can I limit which tools are exposed to my MCP host?

Yes. Set the environment variable `AZURE_AI_SEARCH_MCP_TOOL_GROUPS` to a comma‑separated list of tool groups (e.g., `READ_INDEX,READ_DOCUMENTS`). Use `ALL` to include every tool.