There is no key to obtain and no account to create, because there is no API contract underneath — it drives the recruiting sites themselves. That makes it the fastest way to put job search into an assistant, and it also means the failure mode is a site redesign rather than a rate-limit error. The crawler settings exist for exactly those days.
A job-search server that crawls mainstream recruiting sites and returns listings in one normalised shape. `npx -y mcp-jobs` and nothing else: no registration, no API key, no environment variables. The prompts and the results are in Chinese, aimed at the Chinese job market.
- Search by role, city, salary range and other filters, phrased in ordinary language
- Job detail retrieval from a listing link, so a posting can be analysed without opening it
- Results merged across several sites into one consistent format
- Crawler behaviour you can adjust when a site is slow or you are debugging: `CRAWLER_HEADLESS`, `CRAWLER_TIMEOUT`, `CRAWLER_VIEWPORT_WIDTH`, `CRAWLER_VIEWPORT_HEIGHT`, `CRAWLER_USER_AGENT`, `CRAWLER_DEBUG` — every one of them optional
Node.js 16.0.0 or newer, and `npx -y mcp-jobs` in the client config with the environment left empty. The package is `mcp-jobs`. Because the data comes from crawling rather than an API, the project asks you to respect each site's terms and keep request rates reasonable — and site changes, not your configuration, are what will break a query.
One command — npx -y mcp-jobs