ai-gateway

AI Gateway tier in Azure API Management

One gateway for every model and tool. AI Gateway tier in Azure API Management is the AI-focused tier of Azure API Management. It gives developers one familiar way to call model APIs and MCP tools, while platform teams enforce security, policy, networking, and observability.

• Govern model APIs — Microsoft Foundry, OpenAI, Anthropic, Google Vertex, and more, while developers keep the SDKs they already use.
• Federate MCP tools — combine MCP servers, OpenAPI APIs, and SaaS connectors behind one governed MCP endpoint.
• Enforce AI policies — access, content safety, token, and rate controls, globally or per model and tool.
• Observe AI traffic — token usage, latency, errors, and policy outcomes in dashboards and logs.
• Run on a trusted platform — the Azure API Management platform, with Private Link, virtual network integration, and managed identity.

AI Gateway is in public preview, available free of charge. See the documentation to get started.

---

Coding-agent plugin

This repository hosts the official plugin for coding agents. It helps developers build applications on top of an existing AI Gateway tier in Azure API Management resources: it discovers the models and MCP tool servers registered in a gateway and integrates them into whatever you're building — call a model over the OpenAI-compatible passthrough, connect MCP tools, or scaffold a runnable agent — all from your coding agent.

The plugin follows the Claude Code plugin spec, the GitHub Copilot / awesome-copilot plugin convention, the Cursor plugin format, and the Gemini CLI extension format, so it installs in any of those coding agents (and other agents that support the plugin standard).

Consumption only. The plugin is read-only against the gateway — it never provisions, updates, or deletes resources. Creating gateways, models, or tools is an administrator task done in the AI Gateway Portal.

What's inside

Component	Invoked as	Purpose
Skill use-ai-gateway	/ai-gateway:use-ai-gateway	Full discover → select → credential → integrate workflow (call a model, connect MCP tools, or scaffold an agent).
Command discover	/ai-gateway:discover	Read-only: list a gateway's models and MCP tool servers and pick the ones that fit your use case.
Command build	/ai-gateway:build	End-to-end: discover, retrieve a credential, and integrate the models/tools into your app.

Install

The plugin is hosted in this repository, so your coding agent can install it directly from Git — no manual file downloads. Pick the tab for your agent:

Claude Code / GitHub Copilot CLI (plugin marketplace protocol):

/plugin marketplace add Azure/ai-gateway
/plugin install ai-gateway@azure

Cursor — add this repository as a team marketplace from Dashboard → Plugins → Add Marketplace → Import from Repo (https://github.com/Azure/ai-gateway), then install ai-gateway from Customize. For local testing, symlink the repo into ~/.cursor/plugins/local/ai-gateway and reload the window.

Gemini CLI:

gemini extensions install https://github.com/Azure/ai-gateway

Usage

Point the plugin at your gateway's ARM resource id (required to list its models and tools) and describe what you want to build:

/ai-gateway:build /subscriptions/<sub>/resourceGroups/<rg>/providers/Microsoft.ApiManagement/service/<gateway> a service that summarizes support tickets

Or explore first, then integrate:

/ai-gateway:discover /subscriptions/<sub>/resourceGroups/<rg>/providers/Microsoft.ApiManagement/service/<gateway>

Security

The generated code authenticates the model passthrough and the MCP tool servers with a gateway API key, read from an environment variable (AI_GATEWAY_API_KEY) and never hardcoded. Scaffolded projects keep .env out of source control. If a key is ever exposed, rotate it in the AI Gateway Portal.

To report a security issue, see SECURITY.md.

Samples

Looking for working examples? The samples catalog curates end-to-end samples and infrastructure that build on the AI Gateway tier in Azure API Management.

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit Contributor License Agreements.