Live ATS Jobs Search & Posting Verification — MCP & API

Give an AI assistant current employer-published job data. Search selected Greenhouse, Lever and Ashby boards, or check whether supplied posting IDs are still present in their official public feeds. Results include source URLs, evidence paths, observation times and explicit coverage gaps.

No LLM, provider key, n8n account, company login or proxy is needed. The Actor performs deterministic source extraction. Your AI client can interpret the results using its own model.

Try it

Leave Operation = Fictional demo and press Start. The built-in example finds one fictional fully remote automation role mentioning n8n. It makes no network requests and produces zero paid query events. The result is clearly marked isDemo: true.

For a live example, set mode: "search", enable useExampleBoards, and set maxJobs: 10. This reads public boards for Anthropic, Palantir and Notion; it is a normal paid query. These examples do not imply affiliation with those companies.

What it solves

• Fresh source data: read the employer's official public API at request time, rather than assuming a search-engine result is current.
• Remote vs Hybrid: use explicit workplaceType. Ashby's isRemote=true can mean Remote or Hybrid, so that flag alone never qualifies a role as fully remote. Greenhouse location text remains separate from structured workplace evidence.
• Posting verification: match IDs in source data. A generic HTTP 200 page is not treated as proof that an individual role is listed.
• Auditable keyword matches: literal words/phrases, short normalized excerpts, and JSON source paths. No invented relevance or hiring-intent scores.
• Known unknowns: blocked, empty, failed and incomplete feeds do not turn into closed-job claims.

This is a per-company tool, not a worldwide job search database. You supply hosted board URLs or opt into the example boards. No background index, monitoring schedule or recurring server is created.

Search example

{
  "mode": "search",
  "boardUrls": ["https://jobs.ashbyhq.com/notion"],
  "keywords": ["engineer"],
  "keywordMode": "any",
  "searchDescription": false,
  "remoteOnly": true,
  "maxJobs": 20
}

Search accepts 1–5 board URLs. Supported hosted domains: boards.greenhouse.io, job-boards.greenhouse.io, jobs.lever.co, jobs.eu.lever.co, jobs.ashbyhq.com. Custom company career sites and Greenhouse EU hosts are outside this version's scope.

Up to 15 literal keywords, 60 characters each. any matches any keyword; all requires every one. Keywords default to title-only matching. Enable searchDescription to inspect description text too. AI does not match the middle of mail; this is word/phrase matching, not semantic AI search.

remoteOnly requires explicit structured Remote. Hybrid and unknown types are excluded, including Greenhouse rows whose location text says Remote without a structured workplace field. This does not determine country, visa or work-from-anywhere eligibility.

Up to 100 matched jobs are returned. The summary reports all observed matches and whether output was truncated. Ordering prioritizes title keyword matches, then stable source IDs; it is not a candidate-fit ranking.

Verification example

{
  "mode": "verify",
  "jobUrls": ["https://job-boards.greenhouse.io/YOUR_BOARD/jobs/YOUR_NUMERIC_ID"]
}

Replace the example with a real posting URL. Supply 1–20 URLs from at most five boards. Queries/grouping use official APIs, not the supplied HTML page. Tracking parameters are dropped for identity; repeated posting URLs are marked duplicate and are not re-fetched individually.

Status	Meaning
listed	Matching ID exists in the source's listed public postings
direct_link_only	Ashby exposes the posting with isListed=false
visible_listing_unspecified	Posting exists in Ashby API, but listing visibility was not supplied
not_in_public_feed	ID absent from a complete, successfully read, nonempty feed
unknown	Source cannot establish presence or absence
duplicate	Same canonical posting already checked in this query

Absence from a feed is not proof of permanent closure or that someone was hired. Presence is not proof of genuine hiring intent. This does not label roles as “ghost jobs.”

Use from an AI assistant via MCP

The Actor is available through Apify's hosted MCP gateway:

Connect this Actor through MCP

Configure that URL in a supported remote MCP client using your own Apify authentication. Alternatively, load this Actor by name through Apify's call-actor tool. It runs a finite job and exits; there is no standalone Standby server or idle-time service to maintain.

Example request to your assistant:

Search these employer board URLs for automation or n8n mentions. Return titles, locations, exact source evidence and URLs. Distinguish Remote from Hybrid and unknown workplace types. Verify selected posting links before I use them. Do not assume a company wants outsourcing or that a published role proves actual hiring intent.

The assistant's own model/subscription cost is separate. No model is called by this Actor.

Outputs

Dataset: job results, verification records and explicit diagnostics. Inspect recordType before processing rows.

Job fields include provider, board, jobId, jobKey, title, locations, workplaceType, workplaceTypeSource, jobUrl, sourceJobUrl, sourceApiUrl, sourcePath, observedAt, matchedKeywords and matchEvidence. Full job descriptions are not copied; matched body text produces short normalized excerpts. Salary text is returned only where directly provided in supported structured source fields, without inference or annual/currency conversion.

Dates preserve their source meaning: Greenhouse first_published and Ashby's last publication time appear with publishedAtMeaning; Lever creation time is createdAt, not an invented posting date. observedAt records this check.

Key-value store: OUTPUT contains the full report, per-source requests, errors, coverage, limitations and billing. SUMMARY contains counts and event totals. A healthy search with no keyword matches returns an explicit no_matches diagnostic, not a fabricated job.

Pricing

$0.01 per completed live query, including Apify platform usage. One search covers up to five boards; one verification batch covers up to 20 URLs / five boards. There is no per-job, separate source or model fee. Repeated successful queries are billed again.

A search is billable when at least one nonempty source feed was fully read, even if no job matches the filters. Verification is billable when at least one unique posting has source-backed presence or absence evidence. Partially successful batches retain source diagnostics and still count as one query. Fictional demo, invalid input and entirely unusable evidence produce zero query events. Upstream rate limits/availability are not guaranteed.

Own-account development tests use your account's usual Apify credits. Other users invoke and pay for the Actor under their own accounts. No external provider API charges are used by this application.

Bounds and data handling

• Source response cap 10 MB; 5,000 postings per board; at most ten Lever pages (250 requested per page). A capped/incomplete feed cannot establish absence.
• Source request timeout up to 15 seconds; total source-fetch budget 45 seconds. No retries, redirects, logins, applications, browser rendering or anti-bot bypass.
• Application logs omit source bodies. Normal Apify input/output storage and account retention apply. Only public job-posting endpoints are read; no applicant profiles, resumes or private ATS records are accessed.
• Employer labels/snippets/links are untrusted data. API availability and schemas can change. Review your rights to reuse upstream job content; the code's licence does not license employer content.

Source documentation: Greenhouse Job Board API, Lever Postings API, Ashby public postings API. Independent product, not affiliated with those providers or example employers.

Development

Node.js 22+. npm ci, npm test, apify run for the fictional demo. npm run test:live reads the three public example boards and verifies a currently observed posting from each, without Apify runs or model calls. Live results are time-specific; fixture tests provide deterministic failure/edge-case coverage.