// the find
Azure-Samples/AI-Gateway
Labs to explore AI Models, MCP servers, and Agents with the AI Gateway powered by Azure API Management and Microsoft Foundry 🚀
A collection of 30+ Jupyter notebook labs showing how to use Azure API Management as a gateway layer for AI workloads — rate limiting, semantic caching, MCP protocol support, multi-model routing, and agent orchestration. Aimed at teams on Azure who want a centralized control plane for their LLM API calls rather than calling model endpoints directly from application code.
Each lab ships with Bicep infrastructure templates and APIM policy XML, so you're not just reading theory — you can deploy the exact configuration to your subscription. The semantic caching lab is genuinely useful: it uses vector similarity to avoid re-hitting the LLM for near-duplicate prompts, which directly reduces costs. MCP coverage is current — the labs include client authorization flows and A2A agent patterns that most enterprise gateway tutorials still ignore. The Copilot Agent Skills in .github/skills let you generate new labs from a prompt, which is a practical shortcut for adapting the patterns to your own scenarios.
The repo is Azure-only and requires Azure API Management, which is expensive (Developer tier starts at ~$50/month, Premium is several hundred) — there's no path here if you're not already committed to Azure. Everything runs through Jupyter notebooks, which means the 'labs' format doesn't translate cleanly into production code you'd actually ship; you're expected to lift and adapt yourself. The multi-model support is shallow on non-Microsoft providers — AWS Bedrock gets one lab, Google Gemini gets one, but the routing intelligence is basic policy XML rather than anything model-aware. Several labs are in a _deprecated folder with no clear guidance on what replaced them or why, which is a maintenance smell.