EvoMap

These guides are available in English.

CloudBase-AI-Toolkit

CloudBase-AI-Toolkit connects AI coding tools to Tencent CloudBase through plugins, Agent Skills, and MCP. For DeepSeek Harness, the related @cloudbase/dsh-plugin package provides an MCP bridge, database query result cards, and deployment previews. You still need your own CloudBase environment and should confirm sensitive actions.

Source repository: TencentCloudBase/CloudBase-AI-Toolkit

Discover cloud workflow assets Check EvoMap integration routes

Unified Cloud Deployment — pixel art illustration

Deploying to the Cloud Without Leaving the Session

The toolkit lets an AI agent log in, inspect and change data, manage cloud functions and hosting, read logs, and redeploy from a conversation. That reduces tool switching, but it also puts infrastructure actions close to generated code. Keep production deploys, permission changes, and destructive data operations behind explicit approval.

Example requests:

Inspect the failed CloudBase function deployment and summarize logs without changing resources.

Draft a deployment plan for this service, including database and permission changes, but do not apply it.

Cloud Functions, Databases, and Hosting, Scaffolded Together

CloudBase provides managed database, storage, authentication, cloud functions, Cloud Run, and hosting capabilities. Skills guide project structure and practices; MCP handles environment and resource operations. The toolkit does not make every generated permission rule safe, and it does not remove the need to validate application behavior.

Cloud Modules Working Together — pixel art illustration

For code-review support, see brooks-lint. For browser-based verification after deployment, review BrowserSkill.

How This Differs From a Generic Deployment Script

A static script executes a fixed sequence. The toolkit exposes environment-aware tools and guidance to the agent, which can inspect state and propose the next action. Use a controlled workflow:

  1. Select the correct China or international CloudBase site and environment.
  2. Inspect existing resources, permissions, and deployment state.
  3. Plan every create, update, and delete action.
  4. Approve sensitive changes individually.
  5. Verify logs, endpoints, auth behavior, and rollback readiness.

Example output contract:

Return environment ID, planned resource changes, permission impact, commands or tools to be used, verification steps, and rollback plan.

This preserves human control while still using the agent for coordinated backend work.

Installing CloudBase-AI-Toolkit and Connecting an Account

Use the official CloudBase DSH plugin guide and choose the runtime surface first.

For the Web profile with an MCP bridge, database query result cards, and deployment previews, use the dedicated installer:

curl -fsSL https://raw.githubusercontent.com/TencentCloudBase/CloudBase-AI-Toolkit/main/dsh-plugin/scripts/install.sh | bash

A manual Web setup must add the package, allow required install scripts, install profile dependencies, rebuild the DSH frontend, and start the Web profile:

dsh plugin --profile web add @cloudbase/dsh-plugin
echo "enable-scripts=true" >> ~/.dsh/profiles/web/.npmrc
(cd ~/.dsh/profiles/web && pnpm install)
pnpm --dir "$(dirname "$(dirname "$(command -v dsh)")")" run build:web
dsh --profile web

For a headless MCP bridge without database query result cards, deployment previews, or a frontend rebuild:

dsh plugin --profile headless add @cloudbase/dsh-plugin
dsh --profile headless "list all mcp__cloudbase__ tools"

If no local tcb login state exists, call mcp__cloudbase__auth with action=start_auth and authMode=device, complete the browser authorization, then call it again with action=set_env to select the intended environment. Begin with a read-only inspection and keep long-lived secrets out of prompts and committed configuration. For persistent context, compare mem9.

Cloud-integration status checked 2026-09-28: DeepSeek Harness upstream status still says developer preview and compatibility-breaking changes are expected. Rebuild and retest the Web profile after updates.

Turn a Verified Cloud Fix Into an EvoMap Asset

CloudBase performs cloud operations; EvoMap separately organizes verified agent assets. Once the environment, resource changes, permissions, verification result, and rollback are approved, a team can explicitly package the repair for reuse. Start with the EvoMap Wiki.

FAQ

Does the toolkit include a CloudBase account?

No. You need your own CloudBase environment and its authorization flow. Keep environment selection and account approval outside the reusable procedure.

Can the agent deploy functions and hosting?

The toolkit exposes capabilities for functions, Cloud Run, hosting, logs, and redeploys. Confirm sensitive changes and verify the result in your environment.

Is the DeepSeek Harness package the same as the top-level MCP package?

No. The toolkit points to a related DSH plugin package, while the top-level project also ships the @cloudbase/cloudbase-mcp package and other integration paths. Use BrowserSkill when a deployed interface also needs browser verification.

Preserve a Verified Cloud Repair

Preserve a cloud repair only after the environment, resource changes, permissions, verification result, and rollback plan are recorded.

Browse EvoMap deployment assets Verify flows with BrowserSkill