Ats Job Feed Recipes MCP Server

ATS job feed recipes

Working code for pulling job postings straight out of company applicant tracking systems — Greenhouse, Lever, Workday, Ashby, Workable, SmartRecruiters, Breezy, Personio, Pinpoint and Rippling — and doing something useful with them.

Every script here runs as-is. Set APIFY_TOKEN and go.

export APIFY_TOKEN=...        # https://console.apify.com/settings/integrations
python python/new_jobs_to_csv.py

Why this exists

Each of these platforms publishes an open JSON endpoint that the company's own careers page reads — no key, no login, no scraping of rendered HTML. The awkward parts are everything around that:

• Every platform has a different shape. Greenhouse nests offices and departments; Workday sends a POST and reports posting age as the string "Posted 30+ Days Ago"; SmartRecruiters caps a response at 100 rows whatever limit you pass; Breezy answers 403 for a board that does not exist, where everyone else uses 404.
• Finding out who is on which platform is the actual work. There is no public directory of Greenhouse or Personio boards. The slug is in a careers URL somewhere, if you can find the careers URL.
• Dates are unreliable. Three of these ten platforms publish no posting date at all, so any "posted in the last N days" filter has to decide what to do with rows whose age is unknown.

The scripts below use Apify Actors that have already dealt with all of that and return one row shape across every platform. You can equally point them at the raw endpoints — the notes at the bottom list them.

Recipes

These Actors bill per row delivered. Every script caps itself at a couple of hundred rows and reads MAX_ROWS if you want more, so running an example does not produce a surprise.

The row shape

Every recipe gets the same fields, whichever platform the job came from:

{
  "provider": "greenhouse",
  "company": "Databricks",
  "companySlug": "databricks",
  "jobId": "7845321",
  "title": "Staff Software Engineer",
  "location": "San Francisco, CA",
  "department": "Engineering",
  "employmentType": "Full-time",
  "remote": false,
  "postedAt": "2026-08-14T09:12:00.000Z",
  "applyUrl": "https://boards.greenhouse.io/databricks/jobs/7845321",
  "salary": null,
  "descriptionText": "...",
  "scrapedAt": "2026-08-19T11:02:41.883Z"
}

companySlug:jobId is stable while a posting is open, which is what makes the diffing recipes work.

What each platform actually publishes

Measured from live boards, not from vendor documentation. Useful before you build on any one of them:

Workday's partial is the "Posted 30+ Days Ago" problem: an exact age becomes a timestamp, a vague one stays null rather than being invented.

Raw endpoints

If you would rather call the platforms directly, these are the public endpoints. All of them answer without a key. Replace {company} with the board slug.

Two things that will bite you if you write your own client:

• SmartRecruiters returns at most 100 postings per response regardless of limit. Page with offset, and take the real total from totalFound rather than counting rows.
• Breezy returns 403, not 404, for a subdomain with no board on it. Treating that as rate limiting and backing off turns a half-hour job into a fourteen-hour one.

Licence

MIT. The Actors these scripts call are commercial and priced per row; the code here is not.