GPT RAG Model Context Protocol Server

Part of GPT‑RAG

The GPT-RAG MCP service deploys a Python MCP server built with FastMCP and Starlette. It is consumed by GPT-RAG through the orchestrator mcp strategy and is not tied to an AutoGen-specific runtime. Documentation on the Model Context Protocol can be found here: Model Context Protocol

Prerequisites

Before deploying the application, you must provision the infrastructure as described in the GPT-RAG repo. This includes creating all necessary Azure resources required to support the application runtime.

Initialize the template:

• Ensure the .azure directory is present in the root
• azd deploy

Deploying the app with a shell script

To deploy using a script, first clone the repository, set the App Configuration endpoint, and then run the deployment script.

PowerShell (Windows)

$env:APP_CONFIG_ENDPOINT = "https://.azconfig.io"
cd gpt-rag-mcp
.\scripts\deploy.ps1

Instructions Post Deployment (Must be executed for MCP Server to work properly)

• Navigate to App Configuration resource in Azure portal
• Update AGENT_STRATEGY variable to mcp
• Update MCP_SERVER_URL variable to /mcp (e.g. https://{container-app-name}.{container-app-region}.azurecontainerapps.io/mcp) or the desired MCP Server URL (if external)
• In Azure Portal, Stop Orchestrator Container App, then restart it
• Open frontend URL in your browser and ask “What tools are available to you?” If the MCP Server is working properly, it will list tools and their functionality.

Start MCP Model Inspector to Test MCP connection

• Run the following command in bash or pwsh

npx @modelcontextprotocol/inspector

• Click on the link displayed in terminal that says “MCP Inspector is up and running at…”
• Plug in your container Application URL (found on container app overview page in Azure portal) followed by /mcp (e.g. https://.eastus.azurecontainerapps.io/mcp)

Deploy Locally

• Create a directory .vscode in your root
• Move launch.json into .vscode
• Update the app configuration resource
• Run VS Code Debugger