WPORT CLI Guide: Let AI Agents Search Jobs and Pull Data Directly in the Terminal

When you tell AI in Cursor “find me PM jobs in Taoyuan,” it usually can only search the web or ask you to copy and paste manually.

If we turn job search into a CLI, an AI Agent can run commands in the terminal, get structured JSON, then continue into resume customization, JD matching, or even writing into your Obsidian notes. That is why WPORT CLI (@wport/cli) exists.

---

Why CLI, not a webpage?

Web UIs are great for human clicking. Command-line interfaces (CLI) are great for programs and AI Agents.

Scenario	Web GUI	WPORT CLI
Humans browsing jobs	Intuitive and polished	Possible, but not the main use
AI Agent auto search	Needs click simulation, unstable	One command, stable
Output for the next Skill	Have to scrape HTML	Native JSON / NDJSON
Scripts or CI	Hard	pipe into jq

WPORT CLI is the terminal interface for the W101 Talent Search Center public API. Public job search and viewing do not require login. Employer-side job management has a separate enterprise subcommand (API Key required; briefly covered at the end).

Version note: The latest npm release is currently 0.5.0, still in the 0.x pre-release stage. API domains and commands may change. Pin a version in automation scripts.

---

Install

Requires Node.js >= 18.17.

npm install -g @wport/cli

After install, type wport in the terminal. If an AI Agent (such as Cursor Agent) has terminal access, it can run these commands for you too.

Verify the environment:

wport doctor

doctor shows the configured API base, locale, connection status, and API behaviors Agents should watch for (for example, sort fields are decided by the server and cannot be overridden by the client).

---

Three-step quick start

1. Search jobs

wport jobs search --keyword "產品經理"

Add location codes to narrow results (location codes can repeat with -l):

wport jobs search --keyword "PM" -l 6001001000

By default the terminal shows a table. If output is piped, it automatically switches to JSON for downstream processing.

2. View a single job

Copy enc_id from search results, then run:

wport jobs view <enc_id>

You can also fetch one field only to reduce tokens the AI reads:

wport jobs view <enc_id> --field job_info.job_title

3. Output JSON for Agents or jq

wport jobs search --keyword backend --output json

If you only need fields an Agent requires, use --minimal (a compact field set):

wport jobs search --keyword backend --minimal --output json

Or choose fields yourself:

wport jobs search --keyword backend --fields enc_id,title --output json

--minimal and --fields trim JSON on the client, so the LLM reads fewer irrelevant fields. Network payload size stays the same. This is designed specifically for Agent workflows.

---

Practical examples for AI Agents

Example A: Search → extract enc_id → batch-read titles

wport jobs search --keyword backend --fields enc_id --output json \
  | jq -r '.data[].enc_id' \
  | wport jobs view - --batch --field job_info.job_title

--batch mode reads multiple enc_id values from stdin and outputs NDJSON line by line. One failure does not stop the whole batch. Good for Agents processing many JDs at once.

Example B: Ask the Agent to run it in Cursor

In Cursor Agent mode, you can say:

Please run wport jobs search --keyword "桃園 產品經理" --minimal --output json in the terminal, turn the results into a Markdown table, and mark the 3 listings best for new grads.

The Agent will call the CLI, get JSON, then continue into resume advice based on your Skill or conversation. Much more reliable than asking it to “find jobs on the website.”

Example C: Advanced query (--json-query)

Some filters (years of experience, remote mode, salary range, and more) are not yet exposed as short flags. Write them into a JSON file and pass it in:

wport jobs search --json-query ./my-search.json --output json

The JSON shape matches the server-side JobSearchDto, which is useful for advanced scripts or Agents that generate query files dynamically.

---

Locale and personal settings

CLI UI strings are fixed. --lang affects the language of API response content (job descriptions and similar):

wport jobs search --keyword backend --lang en-US

Persist preferences:

wport config set locale zh-TW
wport config get

Configurable items: locale, output, timeout_ms. Set the API base via environment variable or flag, not config (for safety):

WPORT_API_BASE=https://api.wport.me wport jobs search --keyword backend
# or one-off
wport jobs search --keyword backend --api https://api.wport.me

---

Enterprise commands (recruiters)

If you manage WPORT jobs for a company, use the wport enterprise subcommand (requires enterprise API Key wpk_live_...; contact the BD team to get one):

wport enterprise login          # store the key securely (does not appear in shell history)
wport enterprise jobs list --status published
wport enterprise jobs create --file new-job.json
wport enterprise talents list --tab applied

Enterprise commands are fully isolated from public wport jobs. Without login, you never touch employer APIs. CI / Agents can also pass the key via environment variable:

WPORT_API_KEY=wpk_live_... wport enterprise jobs list --minimal --output json

---

Exit codes and rate limits

When Agents write scripts, use exit codes to judge results:

Code	Meaning
0	Success
2	Bad arguments
3	Server 4xx
4	Server 5xx / network error
5	Corrupted config

Public job search is limited to about 1200 requests / minute / IP, enough for normal interactive use and single-threaded Agents. Do not hammer it with multi-process crawlers.

---

Why we designed it this way

1. Agent first: --minimal, --fields, --batch, and pipe-friendly JSON defaults all help AI read less noise and do more work.
2. Public and enterprise separated: Job seekers use wport jobs with no account. Company features live under enterprise, with clear key permissions.
3. Scriptable: Chaining with jq, Skills, and MCP toolchains turns “search jobs” into one link in resume customization and application tracking pipelines.

This was also the core demo at Charging Station Vol.3: one wport jobs search, and Taoyuan PM jobs enter the Agent workflow immediately. If you missed the class, this article is the make-up handbook.

---

Next steps

• Install the package: npmjs.com/package/@wport/cli
• Troubleshooting: Run wport doctor first. During 0.x, email yao@wport.me
• Browse jobs on the site: wport.me

We will keep updating the “WPORT features” section with new commands, Agent integration examples, and product design trade-offs. Bookmark this post, and next time you open Cursor, just ask the Agent to run wport.