Azure AI Agent Service MCP Server

This MCP server integrates with Azure AI Foundry to enable connections to your existing Azure AI Agents, utilizing the wide range of models and knowledge tools available within Azure AI Foundry, such as Azure AI Search and Bing Web Grounding.

demo

GitHub watchers

GitHub forks

GitHub stars

Azure AI Community Discord

Features

🤖 Agent Integration - Connect to your existing Azure AI Agents
🔄 Seamless Workflow - Use your agents directly within any MCP client
🛡️ Secure - All connections use your Azure credentials
🧠 Conversation Memory - Each client session maintains isolated conversation history

Tools

connect_agent
- Connect to a specific Azure AI Agent by ID
- Inputs:
  - agent_id (string): The ID of the Azure AI Agent to connect to
  - query (string): The question or request to send to the agent
  - thread_id (string, optional): Thread ID for continuation of conversation
- Returns: Formatted response from the agent
query_default_agent
- Send a query to the default configured agent
- Inputs:
  - query (string): The question or request to send to the agent
  - thread_id (string, optional): Thread ID for continuation of conversation
- Returns: Formatted response from the default agent
list_agents
- List all available Azure AI Agents in your project
- Returns: List of available agents with their IDs and names

Configuration

Setting up Azure

Create Azure AI Agents through Azure AI Foundry
Note your Azure AI Project connection string
Note your agents' IDs (you'll need these to connect to specific agents)
Authenticate using Azure credentials: az login

Environment Variables

This server requires the following environment variables:

# Required
PROJECT_CONNECTION_STRING=your-project-connection-string

# Optional (configure default agent)
DEFAULT_AGENT_ID=your-default-agent-id

Installation

Prerequisites

Python 3.10+
Azure CLI (az) installed and configured
Existing Azure AI Agents with desired capabilities

Python Setup

# Setup environment
python -m venv .venv
.venv\Scripts\activate  # On Windows
source .venv/bin/activate  # On macOS/Linux

# Install dependencies
pip install mcp[cli] azure-identity python-dotenv azure-ai-projects aiohttp

# Run server directly (from ./src/python)
python -m azure_agent_mcp_server

If you prefer using uv:

# Setup environment with uv
uv venv
.venv\Scripts\activate  # On Windows
source .venv/bin/activate  # On macOS/Linux

# Install dependencies
uv add mcp[cli] azure-identity python-dotenv azure-ai-projects aiohttp

# Run server (F)
uv run -m azure_agent_mcp_server

TypeScript Setup

# Navigate to the TypeScript directory
cd src/typescript

# Install dependencies
npm install

# Build the server
npm run build

# Run the server
npm start

Usage with Claude Desktop

To use with Claude Desktop, add the following to your configuration file:

Python Version

{
  "mcpServers": {
    "azure-agent": {
      "command": "uv",
      "args": [
        "--directory",
        "/ABSOLUTE/PATH/TO/PARENT/FOLDER",
        "run",
        "-m",
        "azure_agent_mcp_server"
      ],
      "env": {
        "PROJECT_CONNECTION_STRING": "your-project-connection-string",
        "DEFAULT_AGENT_ID": "your-default-agent-id"
      }
    }
  }
}

If you don't want to use uv, you can use python:

{
  "mcpServers": {
    "azure-agent": {
      "command": "python",
      "args": ["-m", "azure_agent_mcp_server"],
      "cwd": "/ABSOLUTE/PATH/TO/PARENT/FOLDER",
      "env": {
        "PYTHONPATH": "/ABSOLUTE/PATH/TO/PARENT/FOLDER",
        "PROJECT_CONNECTION_STRING": "your-project-connection-string",
        "DEFAULT_AGENT_ID": "your-default-agent-id"
      }
    }
  }
}

TypeScript Version

{
  "mcpServers": {
    "azure-agent": {
      "command": "node",
      "args": ["/ABSOLUTE/PATH/TO/mcp-foundry/src/typescript/build/index.js"],
      "env": {
        "PROJECT_CONNECTION_STRING": "your-project-connection-string",
        "DEFAULT_AGENT_ID": "your-default-agent-id"
      }
    }
  }
}

Usage with Other MCP Clients

This server follows the MCP protocol specification and can be used with any MCP-compatible client. Refer to your client's documentation for specific instructions on how to connect to external MCP servers.

Development Notes

This project follows a polyglot structure with implementations in both Python and TypeScript:

Python Development

Python code is located in the src/python directory
Always activate the virtual environment from the project root
For package installation, ensure you're in the Python directory where pyproject.toml is located

TypeScript Development

TypeScript code is located in the src/typescript directory
Uses ES Modules for modern JavaScript compatibility
Standard npm workflow: npm install → npm run build → npm start

License

This project is licensed under the MIT License. This means you are free to use, modify, and distribute the software, subject to the terms and conditions of the MIT License. For more details, please see the LICENSE file in the project repository.