ats-jobs-mcp
Enables users to discover applicant tracking system job boards from company domains, list open roles, detect hiring changes over time, and generate hiring summaries across multiple ATS platforms.
README
Greenhouse, Lever, Ashby, Workable, Teamtailor Jobs API
One API for six applicant tracking systems: Greenhouse, Lever, Ashby, Workable, Teamtailor and SmartRecruiters.
Give it a company website. It finds the job board itself, returns every open role, and on every run after that returns only what changed — roles opened, roles closed, titles moved, salaries changed — plus derived hiring signals.
{ "domains": ["stripe.com", "spotify.com", "pleo.io"] }
No board tokens. No ATS knowledge required. No browser.
Two ways to run it
| Hosted on Apify | Self-hosted MCP server | |
|---|---|---|
| Setup | none | npx ats-jobs-mcp |
| State between runs | managed | local file |
| Scheduling, retries, storage | included | your own |
| Price | $1 per 1,000 jobs | free |
| Link | apify.com/feedworks/ats-jobs-api | this repo, mcp/ |
The logic is the same code in both. Pick whichever fits.
What you get
| Event | Meaning |
|---|---|
job_indexed |
Baseline: an open role captured on the first run |
job_posted |
A new opening appeared |
job_closed |
An opening disappeared |
job_changed |
Title, location, department or employment type changed |
salary_changed |
Compensation changed |
hiring_accelerating |
This month's postings are 2× or more the recent average |
first_role_in_function |
First ever hire in sales, AI/ML, security, data — a new budget line opened |
hiring_freeze |
Postings dropped to zero after steady hiring |
Every record carries provenance: source URL, HTTP status, fetch time, job count, truncation flag and a confidence value — so an agent can verify before it acts, instead of trusting a claim.
MCP server
npx ats-jobs-mcp
Four tools, usable by any MCP client — Claude Desktop, Cursor, VS Code, your own agent:
| Tool | Question it answers |
|---|---|
find_job_board |
Which ATS does this company use, and what is its board token? |
list_jobs |
What roles are open right now? |
detect_changes |
What changed since the last time I looked? |
hiring_summary |
What is this company hiring for — which functions, which locations? |
Claude Desktop config:
{
"mcpServers": {
"ats-jobs": { "command": "npx", "args": ["-y", "ats-jobs-mcp"] }
}
}
State lives in ~/.ats-jobs-mcp/state.json. Override with ATS_MCP_STATE.
Why it stays cheap to run continuously
Every source is polled with If-None-Match. When nothing changed, the ATS answers 304 with no body.
Measured on the hosted Actor at 512 MB:
| Path | Compute |
|---|---|
Nothing changed (304) |
0.000344 CU per run |
| Baseline, 143 jobs | 0.001386 CU |
That is why watching hundreds of companies every few hours costs cents rather than dollars — and why this project will never open a headless browser to read a job board.
What it deliberately does not do
- No personal data. Recruiter names, emails and phone numbers are stripped at parse time and never stored.
- No headless browser, no proxies. Only public ATS endpoints that answer with JSON.
- No false layoffs. If a source response may be truncated by paging,
job_closedevents are suppressed — a paging artefact must never look like a company shutting down hiring.
Supported sources
greenhouse · lever · ashby · workable · teamtailor · smartrecruiters
Detection order: a link on the company careers page → the board token guessed from the domain → Workable short links resolved through their redirect.
Companies that cannot be resolved return a source_unresolved record explaining why, instead of failing silently.
Recruitee and Personio are not listed because they are not verified working. Personio currently answers 429 to non-browser clients; Recruitee endpoints did not resolve in testing. This project does not claim coverage it cannot deliver.
Repo layout
src/ Apify Actor: adapters, diff engine, derived signals
mcp/ MCP server — the same engine, self-hosted
tools/ discovery utilities (domain → ATS)
.actor/ Apify manifests and schemas
MIT licensed. Contributions welcome, especially new ATS adapters — the interface is one url() and one parse() function per system, see src/adapters/index.js.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。