Why Serverless Is the Best Way to Deploy Enterprise MCP Servers

MCP protocol defines a unified standard for AI large models to interact with external services such as databases and tool APIs, replacing complex JSON interfaces like Function Calling with a standardized description that models can understand directly.

MCP Protocol: A Universal Language for AI Models and External Services

By standardizing service capabilities, MCP enables natural‑language commands to drive applications (e.g., Blender‑MCP for 3D modeling) and is poised to become the factual standard for AI toolchains, comparable to HTTP for the web.

Enterprise‑Level MCP Server Deployment: Reality vs. Ideals

Low deployment efficiency : MCP Server is lightweight and can start with a single command, making traditional cloud VM deployment inefficient.

High resource management cost : Traffic spikes require costly, slow‑to‑scale VM resources, leading to waste during idle periods.

Security and compliance risks : Access to sensitive data without fine‑grained permission controls can be dangerous.

Complex network configuration : Bridging private enterprise networks and public AI model access often involves cumbersome proxy and firewall setups.

Serverless Hosting: The Key to Solving These Pain Points

Extremely low cost : Pay‑as‑you‑go based on function invocations and resource usage; costs drop by over 50% compared to monthly VM rentals for bursty AI tasks.

Elastic scaling : Functions start in milliseconds and auto‑scale with traffic, preventing service crashes during spikes.

Out‑of‑the‑box security and compliance : Fine‑grained permission policies restrict data access, VPC tunnels enable secure private‑network calls, and automatic audit logs satisfy GDPR and similar regulations.

Efficient development and deployment : One‑click deployment via FunctionGraph’s application center or open‑source IaC tools, with versioning and instant rollback capabilities.

One‑Click Deployment of a Popular MCP Server

Steps to deploy the official Fetch MCP Server template:

Open Huawei Cloud FunctionGraph, go to the Application Center, and click “Create Application”.

Select the “MCP Server” template and click “Use Template”.

Leave the agency field empty or select fgs-app-adminagency, choose an existing APIG instance, set the runtime to uvx, and configure the MCP service as shown below.

Confirm and create the application.

Copy the generated invocation URL; for production, bind a custom domain.

Test with a client (e.g., Cherry Studio) using SSE mode and the copied URL.

{
  "mcpServers": {
    "fetch": {
      "command": "uvx",
      "args": [
        "mcp-server-fetch"
      ]
    }
  }
}

Enabling MCP allows the AI model to fetch external documentation (e.g., Huawei Cloud FunctionGraph docs) and answer queries accurately, whereas without MCP the model cannot retrieve up‑to‑date information.

The MCP protocol is reshaping how AI connects to the real world, and serverless hosting removes cost, scalability, and security barriers, letting developers focus on creating value.