How to Convert Alibaba HSF Services to MCP Servers Without Code Changes
This article explains how Alibaba engineers use the Higress AI gateway to offload MCP protocol and transform large‑scale HSF RPC services into MCP servers without modifying application code, covering challenges, required components, step‑by‑step migration, and practical insights for AI‑native infrastructure.
MCP standardizes resource access and multi‑agent interoperability, and its ecosystem is rapidly expanding with thousands of MCP servers and components. Alibaba’s internal practice demonstrates converting HSF services to MCP servers using the Higress AI gateway, enabling protocol offloading without code changes and preserving flexibility for AI‑native applications.
Background
HSF, built on Apache Dubbo, powers Alibaba’s massive microservice ecosystem with millions of production instances and a robust registry and configuration center (the basis for Nacos). Higress, an open‑source AI gateway, handles cross‑VPC communication, offers high performance, and has become the de‑facto gateway for AI workloads.
Why Host via Higress Instead of Native SDK?
MCP specifications evolve quickly; new SPEC releases may break forward compatibility, making SDK upgrades costly and risky.
Current MCP SDKs are immature and not production‑ready, requiring extensive code changes for migration.
Lack of systematic best practices in the MCP ecosystem increases learning overhead.
Hosting with Higress provides protocol conversion, unified authentication, traffic routing, parameter mapping, and security auditing, allowing developers to expose services as MCP servers without deep protocol knowledge.
Key Components
Higress Gateway: Handles MCP traffic and provides cross‑cutting capabilities.
MCP Console: Platform for creating and managing MCP servers, tools, and prompts.
MCP Registry: Service discovery for MCP servers, implemented by the HSF registry.
MCP Metadata Center: Stores prompts, server metadata, and tool definitions, backed by the HSF configuration center.
Migration Steps
step1
Open the appropriate HSFOPS backend and select the MCP sidebar.
step2
Choose the HSF application you own, then specify the service name and method.
step3
Fill in the //TODO fields with the method’s input parameter field names and descriptions; the request schema is generated automatically.
step4
Access the MCP server via SSE using the endpoint http://{MCP endpoint prefix}/{applicationName}/sse and configure the cursor as follows:
{
"mcpServers": {
"{applicationName}": {
"url": "http://{MCP endpoint prefix}/{applicationName}/sse"
}
}
}AI Infra Perspective
MCP is not a silver bullet; its protocol maturity is insufficient for production‑grade AI workloads. Hosting via Higress, which offloads MCP, offers a balanced solution that avoids vendor lock‑in while allowing rapid prototyping and iterative improvement.
Conclusion
The guide provides a practical, no‑code‑change path for large‑scale HSF services to join the MCP ecosystem, accelerating AI‑native application development while acknowledging MCP’s current limitations.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
Alibaba Cloud Developer
Alibaba's official tech channel, featuring all of its technology innovations.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
