cloudbase

name	cloudbase
description	Use when building, deploying, or debugging Tencent CloudBase (腾讯云开发 / TCB) apps in Trae — Web, WeChat Mini Program, cloud functions, CloudRun, auth, NoSQL/PostgreSQL, storage, and built-in AI. Prefer CloudBase MCP tools first, then load published CloudBase skills for scenario details.

CloudBase (Tencent Cloud Development)

Description

Guide Trae to develop on Tencent CloudBase with the correct order of operations: environment binding → MCP management tools → scenario skill → implementation → review/deploy.

CloudBase covers Web / H5, WeChat Mini Program, cloud functions, CloudRun, document DB / PostgreSQL / MySQL, cloud storage, auth, and built-in AI models.

Usage Scenario

Use this skill when the user:

• Mentions CloudBase, 云开发, TCB, 腾讯云开发, or WeChat Mini Program cloud development
• Asks to scaffold, deploy, or debug a CloudBase Web / Mini Program / CloudRun app
• Needs auth, database, storage, cloud functions, or CloudBase AI model integration
• Wants Trae wired to @cloudbase/cloudbase-mcp

Do not use for non-CloudBase backends or pure frontend work with no CloudBase dependency.

Preconditions

• Trae can run terminal commands and edit project files
• Prefer enabling CloudBase MCP (see below) before management API work
• User should have a Tencent Cloud / CloudBase account (browser login is prompted by MCP)

Instructions

1. Confirm scenario — Web, WeChat Mini Program, CloudRun, database-only, auth-only, or AI model. State which scenario you will follow.
2. Ensure CloudBase MCP is available in Trae
   • Settings → MCP → Add → Manual:

{
  "mcpServers": {
    "cloudbase-mcp": {
      "command": "npx",
      "args": ["-y", "@cloudbase/cloudbase-mcp@latest"],
      "env": {}
    }
  }
}

• On first use, complete browser login / environment selection.

3. Bind environment explicitly
   • Call envQuery (or equivalent) to resolve the canonical EnvId.
   • Never pass aliases/nicknames directly into SDK init or deploy configs — resolve to full EnvId first.
4. Prefer MCP tools for management work
   • Auth providers, DB schema, functions, hosting, CloudRun, storage, security rules: use MCP first.
   • Inspect tool schemas before calling; do not invent parameters.
5. Load the matching published CloudBase skill before coding
   • Main entry: https://github.com/TencentCloudBase/skills (or npx skills add tencentcloudbase/cloudbase-skills)
   • Common ids: web-development, miniprogram-development, auth-tool-cloudbase, auth-web-cloudbase, cloud-functions, cloudrun-development, postgresql-development-cloudbase, ui-design, cloudbase-platform
   • Read the skill fully before writing application code.
6. Implementation order
   • Resource prep via MCP (providers, collections/tables, permissions) → then frontend/backend code → local verify → deploy.
7. Close-out
   • Run a CloudBase-aware review when available (cloudbase-code-review skill).
   • Report EnvId, resources touched, and preview/deploy URLs.

Examples

Example 1 — Web app with username/password auth

User: “用云开发做一个带登录的 Web 管理后台”

Agent should:

1. Enable CloudBase MCP and bind EnvId
2. Use auth MCP tools to enable username/password (and publishable key) before writing login UI
3. Follow auth-tool-cloudbase then auth-web-cloudbase / web-development
4. Deploy static hosting only after auth providers are confirmed

Example 2 — WeChat Mini Program CRUD

User: “小程序云开发读写文档数据库”

Agent should:

1. Confirm Mini Program + CloudBase scenario
2. Use NoSQL MCP tools / cloudbase-document-database-in-wechat-miniprogram guidance
3. Avoid Web SDK auth patterns in Mini Program code

References

• Docs: https://docs.cloudbase.net/ai/cloudbase-ai-toolkit/ai-agent-plugins
• Toolkit source: https://github.com/TencentCloudBase/CloudBase-AI-Toolkit
• Open Plugin / skills package: https://github.com/TencentCloudBase/cloudbase-plugin
• Published skills: https://github.com/TencentCloudBase/skills

Constraints

• Do not invent CloudBase API paths or MCP tool arguments
• Do not expose API keys / service_role credentials in frontend code
• After 2–3 failed attempts on the same path, stop and reroute (env, auth domain, permission model, or skill)