Turn Existing OpenAPI Specs into AI‑Ready MCP Tools with Higress

Background – Anthropic released the Model Context Protocol (MCP) in November 2024 to provide a unified interaction model that bridges large language models (LLMs) with heterogeneous data sources and tools, eliminating data silos.

Problem – Enterprises often have many existing OpenAPI specifications that are not directly callable by AI assistants. The challenge is to transform these OpenAPI definitions into MCP tools that AI platforms like Dify can invoke.

Solution Overview – Use the Higress cloud‑native gateway to (1) convert OpenAPI schemas to MCP configurations, (2) route requests to the original APIs, (3) implement dual authentication, (4) connect the MCP tools to AI assistants, and (5) fine‑tune prompts for optimal performance.

Step‑by‑Step Implementation

Convert OpenAPI Schema to MCP – Extract key information (paths, methods, parameters, responses) from the OpenAPI JSON/YAML and generate an MCP YAML using the openapi‑to‑mcp tool provided by Higress.

Configure Higress Routing – Create a ConfigMap with global MCP server settings, add a service source for each legacy API, and define a route that injects the generated MCP YAML.

Dual Authentication –

Higress ↔︎ Backend API authentication (e.g., API keys, OAuth).

User ↔︎ Higress SSE endpoint authentication via Consumer Management (KeyAuth example).

Expose MCP Tool in Dify – Deploy Dify (or use the Compute Nest one‑click deployment), install the “SSE Discovery and Call MCP Tool” plugin, paste the MCP server configuration, and create an agent that calls the MCP endpoint with a chosen model (e.g., QWEN‑MAX).

Prompt Optimization – Use Go templates and GJSON expressions in Higress to refine request/response payloads, and manually adjust prompts if the model misinterprets the MCP schema.

Example: Gaode Maps API – The article walks through converting a Gaode Maps OpenAPI definition into an MCP tool, configuring routing, adding required headers (e.g., Authorization: Bearer xxx), and invoking the tool from Dify to retrieve geographic data.

Key Configuration Snippets

{
  "amap-maps": {
    "headers": {
      "Authorization": "Bearer xxx"
    },
    "transport": "sse",
    "url": "http://12xx.94:8080/amap-maps/sse"
  }
}

Images

Conclusion – By leveraging MCP and Higress, organizations can instantly expose legacy APIs to AI assistants, achieving true “everything‑connected” AI applications across devices, cloud resources, and third‑party services.