As AI agents become more common, exposing Model Context Protocol (MCP) servers reliably is a growing concern. This post shares a hands-on experience of taking an MCP SSE demo from a direct Azure App Service deployment to a production-style setup behind Azure API Management (APIM). The author first validated the backend with MCP Inspector, then tackled the gateway layer. Key challenges include configuring SSE transport correctly in APIM, managing long-lived connections, and ensuring policies don't break streaming responses. The post offers practical insights for developers who need to scale MCP endpoints across multiple services or add authentication and rate limiting at the gateway. While the source is a personal blog, the pattern is directly applicable to enterprise Azure environments. For teams building AI tooling, this is a useful reference for avoiding common integration mistakes.
A practical walkthrough of exposing an MCP SSE server via Azure APIM, covering transport configuration, policy setup, and common pitfalls.