feat: unified server mode — single process serving multiple services

[filhocf]

Problem

The current architecture requires one process per cloud service (mcp-server-ecs, mcp-server-vpc, etc.). With 117+ entry points, running even 13 services means 13 Python processes, each loading the full Huawei SDK (~200MB+ total RAM).

Proposed Solution

Add a unified server mode where a single process loads multiple OpenAPI specs and registers all tools with service-prefixed names (e.g., ecs_ListServers, vpc_ListVpcs).

Usage

export HUAWEI_SERVICES="ecs,vpc,cce,elb,nat,evs,iam"
mcp-server-unified -t stdio

Benefits

Approach	Processes	RAM	Startup
Current (1 per service)	13+	~2GB+	Slow
Unified	1	~200MB	Fast

Design

• New entry point mcp-server-unified
• HUAWEI_SERVICES env var: comma-separated list of service codes
• Each tool prefixed with service code to avoid name collisions
• call_tool routes to correct OpenAPI spec based on prefix
• Compatible with multi-tenant (feat: multi-tenant support via HUAWEI_TENANTS_FILE #141)

Implementation will be available at: https://github.com/filhocf/hcso-mcp-server