How Higress AI Gateway Turns Business APIs into an MCP Server

Background

Previous MCP Server tutorials required writing Java or TypeScript code for each service, similar to developing a regular business API and adding tool annotations. The goal is to expose a REST API as an MCP Server without writing new server code.

Why use Higress AI Gateway

Higress is an open‑source cloud‑native API gateway built on Istio and Envoy. Its MCP Server capability allows declarative conversion of any REST API into an MCP Server, eliminating the need for custom MCP server code.

Setup

Create a Docker network for MCP services: docker network create mcp Pull and run the all‑in‑one Higress image. The example maps ports 8001, 8080, 8443 to host ports 9001, 9080, 9443 to avoid conflicts:

mkdir higress && cd higress
# Pull the latest Higress all‑in‑one image
docker pull higress-registry.cn-hangzhou.cr.aliyuncs.com/higress/all-in-one:latest
docker run -d --rm --name higress-ai -v ${PWD}:/data \
    -p 9001:8001 -p 9080:8080 -p 9443:8443 --network mcp \
    higress-registry.cn-hangzhou.cr.aliyuncs.com/higress/all-in-one:latest

Pull a Redis image for state storage:

docker run -d --rm --name higress-redis -p 6379:6379 --network mcp higress-registry.cn-hangzhou.cr.aliyuncs.com/higress/redis-stack-server:7.4.0-v3

Configure MCP Server in Higress

Develop a simple SpringBoot endpoint (e.g., a GET /test1 returning weather data).

In the Higress console, create a service source that points to the SpringBoot service.

Create an MCP service and add a tool definition. Example YAML:

server:
  description: "获取指定城市的天气"
  name: "mcp-weather"
tools:
- args:
  - description: "城市名称"
    name: "name"
    position: "query"
    required: true
    type: "string"
  description: "根据指定城市获取指定的天气信息."
  name: "get-weather"
  requestTemplate:
    argsToUrlParam: true
    method: "GET"
    url: "http://192.168.4.3:8001/test1"
  responseTemplate:
    body: |-
      # 天气信息
      {{.data}}

Saving the configuration creates an MCP Server that forwards AI Agent requests to the original REST endpoint without changing backend code.

Request template options

The gateway supports four mutually exclusive ways to map tool arguments to the downstream request:

body : Manually construct the request body using a template.

argsToJsonBody : Set argsToJsonBody: true to send all unspecified arguments as a JSON object in the body and automatically add Content‑Type: application/json; charset=utf-8.

argsToUrlParam : Set argsToUrlParam: true to encode unspecified arguments as URL query parameters.

argsToFormBody : Set argsToFormBody: true to encode arguments as application/x-www-form-urlencoded in the body.

Only one option may be used per tool; configuring more than one results in a validation error.

Additional fields prependBody and appendBody allow preserving the original JSON response while adding explanatory or usage information, useful when converting OpenAPI specifications to MCP tools.

Additional features of Higress MCP Server

Unified authentication and authorization for secure AI tool calls.

Fine‑grained rate limiting.

Comprehensive audit logs of tool invocations.

Observability for performance and health monitoring.

Support for multiple MCP protocol versions (e.g., 20241105 and 20250326) and transport modes such as POST+SSE, Streamable HTTP, and native WebSocket.

With these steps, developers can expose existing REST APIs as MCP Servers in minutes, enabling AI agents to access databases, external services, and other data sources without additional code.