CloudBase CLI V3 is here: a command-line tool redesigned for AI agents

We're pleased to announce the official release of CloudBase CLI V3, a CloudBase command-line tool redesigned for AI agents.

This version adds 15 top-level command modules, covering environment management, database, user permissions, routing, domains, logs, AI agents and more. Now every operation of a cloud project, from creation to launch, can be done in the terminal — no console needed.

What V3 changed​

• Filled in the large number of missing commands, so nearly every CloudBase operation can be done in the terminal;
• Every command has -h / --help self-explanation, letting AI agents discover and understand command functions on their own;
• Added the tcb docs command, solving the problem that CloudBase docs were hard for agents to read;
• Full --json output mode, making all command returns machine-readable structured data.

Why CLI V3​

We've observed more and more developers using AI tools like Claude Code, Cursor and CodeBuddy to operate CloudBase — creating environments, deploying functions, configuring routes, handing the whole flow to an agent.

But the traditional GUI console isn't agent-friendly: agents can't click buttons, can't read visual feedback on a page, can't handle popup confirmations. Every time they hit an operation the CLI doesn't cover, the agent has to stop and wait for a human.

A CLI is a design better suited to agents. Command-line input and output are structured text that agents can parse and process directly, no visual understanding needed; commands can be composed and scripted, a natural fit for multi-step automation workflows.

CloudBase CLI V3 solves exactly this: fully redesigned for AI agents, so the agent can run the whole way.

15 new commands: full coverage from environment creation to launch​

Existing commands were also systematically reorganized: command style migrated from colon-separated (tcb functions:deploy) to space-separated (tcb fn deploy), more in line with CLI conventions. This means operations that in the V2 era required switching between "CLI + console" can now all be done in the terminal.

--help: self-explaining commands​

Every command and subcommand in CLI V3 has -h / --help built in, outputting the full usage, parameter list and examples.

Facing an unfamiliar command, an agent doesn't need to check external docs — tcb <command> --help gives structured usage directly:

tcb fn deploy --help

For an agent, this is a zero-cost command discovery mechanism: tcb --help to see the top-level commands, then tcb <command> --help for specifics, exploring layer by layer with no external knowledge needed.

tcb docs: a docs system for agents​

When developing against CloudBase, agents often struggle to find the right docs, and search-engine results are frequently outdated or imprecise.

tcb docs lets agents query official CloudBase docs directly in the terminal, no search engine needed:

# list all doc modules
tcb docs list

# read a doc by path
tcb docs read 云函数
tcb docs read https://docs.cloudbase.net/cloud-function/faq

# full-text keyword search
tcb docs search 微信支付

For an agent, this forms an autonomous troubleshooting loop: hit unknown knowledge → tcb docs search → confirm the content → execute the next step. No human intervention, no external search.

tcb docs works without login.

--json output mode: machine-readable output​

All V3 subcommands support the global --json parameter. With it, output has no progress bars, no color codes, no human-readable decoration — the agent gets pure structured data.

On success:

{"currentEnvId":"luke-personal-test-new-8d0d90f5f"}

On failure:

{"error":{"code":"FUNCTION_NOT_FOUND","message":"函数 ai-reply 不存在","exit_code":4}}

Combined with V3's six structured exit codes, an agent can route responses by exit code:

Failure is no longer a vague error text but a deterministic signal. Agents don't need semantic understanding to guess the next step.

In practice: Claude Code + CLI V3 building an intelligent support ticket system​

We built an intelligent support ticket system with Claude Code + CloudBase CLI V3 to verify V3's real capability boundary. The project covers a database, four cloud functions, HTTP routes, user and permission management, a custom domain and an SSL certificate — the full deployment. The whole flow ran through tcb commands, zero console operations.

Step 1: create and activate the environment

tcb env create --alias "ticket-system-prod" --package baas_personal --yes --json
tcb env use luke-personal-test-new-8d0d90f5f --json

Returns { "currentEnvId": "luke-personal-test-new-8d0d90f5f" }. After setting the global default environment, subsequent commands don't need --env-id. V3's environment priority: global config < cloudbaserc.json < -e CLI parameter.

Step 2: initialize the database: the ticket system needs three collections — tickets, ticket_history, user_roles — with indexes created via tcb db nosql execute.

Step 3: deploy cloud functions: four cloud functions (three event functions, one HTTP function):

tcb fn deploy submit-ticket --json
tcb fn deploy assign-ticket --json
tcb fn deploy reply-ticket --json
tcb fn deploy ai-reply --dir . --yes --json

Step 4: configure security rules and user permissions: database security rules, create an ops account, assign the admin role — the agent closes the loop with five commands.

Step 5: configure HTTP routes: tcb routes add configures the /api/ai-reply route, tcb routes list --json outputs the structured route table.

Step 6: bind the custom domain: look up the certificate, bind the domain, create the CNAME record — three steps in one workflow context. tcb api brings the whole Tencent Cloud ecosystem into one entry point.

Step 7: deploy the front end: cd frontend && tcb app deploy --yes, then open the link to see the published web app.

Summary: all CLI, zero console​

In the past, AI-assisted development stopped at code generation. AI wrote the code, but deployment and ops still needed a human to open the console manually.

V3's design goal is to make tcb an interface agents can fully control. --json machine-readable output, structured exit codes, the tcb api unified entry, tcb docs self-service documentation — the agent can take over the full chain from environment creation to project launch.

Next time you have an AI agent build a cloud project, try not opening the console at all and see how far it gets.

How to use the V3 CLI​

V3's CLI is already on npm:

npm install -g @cloudbase/cli
tcb login

Related docs:

• CloudBase CLI docs: https://docs.cloudbase.net/cli-v1/install
• V3 migration guide: https://docs.cloudbase.net/cli-v1/migrate-v3
• CloudBase MCP: https://docs.cloudbase.net/ai/cloudbase-ai-toolkit/
• cloudbase-skills: https://github.com/TencentCloudBase/cloudbase-skills