
ATS Jobs
io.github.groundtruthtoolsv1.1.0更新於 Oct 7, 2026
Open jobs read from a company's own careers system: Greenhouse, Ashby, Lever.
概覽
直接從企業招募系統(Greenhouse、Ashby、Lever)讀取職缺,並可跨公司比較。
- 功能
- 提供三個工具:find_company_board 依公司名稱查出其使用的招募系統與看板名稱;list_open_jobs 回傳某公司的所有職缺,包含職稱、部門、地點、僱用類型、工作型態、張貼日期、雇主有公布時解析出的薪資區間,以及指向雇主原始徵才頁面的 verify_url;compare_companies 以同一組欄位回傳多家公司,方便直接比較。薪資會解析成數值區間與幣別,並以 salary_text 保留雇主原文。不會回傳招募人員姓名,職缺描述中的電子郵件與電話會被移除。
- 適用情境
- 當你需要來自雇主自家徵才頁面的最新職缺,而不是人力銀行聚合索引時使用;也適合比較採用不同招募系統的多家公司招募狀況,或研究雇主公開的薪資區間。
- 執行需求
- 未宣告需要帳號、API 金鑰或環境變數。可作為遠端 Streamable HTTP 端點使用,也可用 pip 安裝 ats-jobs-mcp 後在本機以 stdio 執行。需要能連線各招募看板的 API。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 ATS Jobs,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
其他 MCP 客戶端
把它新增到你客戶端的 mcpServers 設定中。
{
"mcpServers": {
"ats-jobs-mcp": {
"type": "http",
"url": "https://ats-jobs-api.groundtruthtools.workers.dev/mcp"
}
}
}README
ats-jobs-mcp
An MCP server for open job postings, read first-hand from the applicant-tracking system behind a company's careers page. Ask your assistant what Anthropic is hiring for and get the answer from Anthropic's own board, not from a job aggregator's index.
Built on the job board APIs that Greenhouse, Ashby and Lever publish so third parties can build job sites. No scraping, no API key, no account, no personal data.
Why this exists
Those APIs are open, documented, and useless unless you already know the
company's board name: that Anthropic's is anthropic, Ramp's is ramp,
Mistral's is mistral.ai.
None of the vendors publishes a directory of its customers, and none could,
since no vendor knows about the others. Checked 2026-08-24: sitemap.xml on
job-boards.greenhouse.io, boards.greenhouse.io, jobs.lever.co and
apply.workable.com all 404 or redirect.
So this ships one: 7,479 company boards, every one confirmed against the vendor's own API, each carrying the date it was confirmed.
Three tools
find_company_board — which system is a company on?
Give it a company name. Get the system and the board name, plus how many roles were open when the board was last confirmed.
list_open_jobs — every open role at one company
Title, department, location, employment type, workplace type, posted date, a
parsed salary range where the employer publishes one, and a verify_url
pointing at the employer's own advert.
compare_companies — several companies, one schema
The thing no single vendor can do. Companies on Greenhouse and companies on Ashby come back with the same field names, so they can be compared directly. One company failing does not lose the others.
Salaries, parsed, and refused when unclear
Ashby prints pay as "$211.4K - $290.6K". Greenhouse prints it in cents,
sometimes as several bands at once. Neither is filterable, so both are turned
into numbers:
The parser refuses more often than it guesses, deliberately:
$50 - $80 / hourgives no salary. An hourly rate is a different quantity and reporting it as annual would be a confident lie.600 000 krgives no salary.kris Swedish, Norwegian and Danish.0.5% - 1.75% - Offers Equitygives no salary. Equity is not pay.- Bands in two currencies are never spanned into a range that exists nowhere.
salary_text always keeps what the employer actually wrote, so nothing is
lost when the parser declines.
Every parsed figure is tested against the employer's own structured
baseSalary block, the one published for Google to index — not against a
number someone typed into a test.
No personal data
Recruiter names are never returned. Email addresses and phone numbers are removed from job descriptions, because descriptions genuinely carry them and a column of work addresses is a marketing list rather than a job feed.
Unlisted Ashby postings are dropped, because Ashby's documentation says plainly that they should not be shown publicly.
Coverage, stated honestly
Lever is readable but has almost no boards in the directory yet, so name it
directly (lever:spotify) or pass its careers URL. Workable has a confirmed
endpoint but no adapter yet, so it is not listed at all: a directory entry for
something nothing can read is a broken promise.
Use it with no install at all
The server runs remotely. Point any MCP client that speaks Streamable HTTP at the URL and there is nothing to download:
Checked 2026-09-13: of the job board servers in the official MCP registry, this is the only one with a remote transport. Every other one, including the PyPI build of this one, has to be installed first.
The same thing answers plain HTTP, for agents that do not speak MCP:
GET /openapi.jsonfor a tool-calling agentGET /llms.txtfor a model reading the service in proseGET /v1/jobs?company=stripe&limit=5
Or install it locally
Then add to your MCP client configuration:
Relation to the hosted version
The same parsing runs as an Apify Actor at apify.com/groundtruth/ats-job-listings, which adds scheduling and an incremental mode that returns only postings you have not been sent before. This server is free, runs on your machine, and takes no cut of anything.
tests/test_core_parity.py fails if the two copies of the parsing logic drift
apart.
Licence
MIT. Job postings belong to the employers who published them.
來源:README.md,提交 2b04fe8
工具
0版本歷史
1- v1.1.0最新Sep 16, 2026
