jobdrop 2.0.2

20-source job scraper library + MCP server. LinkedIn, Indeed, Glassdoor, Google, ZipRecruiter, Wellfound, Hiring Cafe, Greenhouse, USAJobs, Adzuna, Jooble, Findwork, The Muse, Insight Global, Clearance Jobs, Kforce, CollabWork, Naukri, Bayt, BDJobs.

jobdrop

A multi-source job scraper. Hits 20 job boards in one call, normalizes the results into a pandas DataFrame, and ships with anti-bot bypasses for the boards that block standard scrapers.

What's in here

17 sources

site_name	Source	Mechanism
linkedin	LinkedIn	Public listing scrape with optional detail-page enrichment
indeed	Indeed	GraphQL with the Int! radius fix + per-company cap + paginate-until-quota
glassdoor	Glassdoor	selenium-driverless headless to defeat Cloudflare 403; in-page GraphQL fetch
google	Google Jobs	selenium-driverless headless against udm=8; SERP DOM walk
zip_recruiter	ZipRecruiter	curl_cffi + safari17_2_ios TLS impersonation against the web HTML endpoint
bayt	Bayt	Public scrape
naukri	Naukri	Public scrape
bdjobs	BDJobs	Public scrape
usajobs	USAJobs.gov	Federal public API
adzuna	Adzuna	Public API
jooble	Jooble	Public API
findwork	Findwork.dev	Public API
the_muse	The Muse	Public API
insight_global	Insight Global staffing	Server-rendered HTML scrape with hidden JSON blob per result
clearance_jobs	ClearanceJobs (DHI)	Public JSON API + parallel detail-page enrichment for full JD, salary, type, remote bool
kforce	Kforce staffing	Direct Azure Cognitive Search calls (bypasses Imperva on the public host)
greenhouse	Greenhouse-hosted boards	Google site: dorks via selenium-driverless → public Greenhouse API; 3-layer staleness filter

Anti-bot solved

• Google — selenium-driverless cold-start headless. Defeats the 2026 CAPTCHA wall that takes out Playwright / undetected-chromedriver / nodriver / patchright.
• Glassdoor — selenium-driverless rewrite to bypass Cloudflare 403; URL-encoded location, partial-GraphQL-error tolerance.
• ZipRecruiter — curl_cffi + safari17_2_ios against the web HTML endpoint. The iOS-app API is dead behind Cloudflare.
• Kforce — bypasses Imperva on the public host by calling the Azure Cognitive Search backend directly.
• Greenhouse — uses the same selenium-driverless infrastructure as Google for site: dorks across all greenhouse-hosted boards.

Other tightening

• LinkedIn — salary extraction from description body, optional per-company cap, parallel detail fetches.
• Indeed — fixed radius=25 default after Indeed promoted the GraphQL field to Int!; per-company cap to surface diverse employers; pagination loop fixed.
• ClearanceJobs — search API gives a 200-char preview; this fork parallel-fetches /api/v1/jobs/{id} so you get the full JD, salary range, structured job_type, and authoritative remote bool.
• Greenhouse — three layers of stale-protection (404 drop / past application_deadline / first_published age with 90-day default that respects hours_old).

Bundled credentials

API keys for USAJobs, Adzuna, Jooble, Findwork, and The Muse are baked into a positional resolver (jobdrop/_defaults.py) so the new sources work without environment setup. User-set env vars still win via setdefault semantics.

Installation

As a Python library

pip install -U jobdrop

Python ≥ 3.10 required.

As an MCP server (Claude Desktop / Claude Code / Cursor / Cline)

Install the binary once with uv tool install (or pipx install):

uv tool install "jobdrop[mcp]"
# or:  pipx install "jobdrop[mcp]"

Then add to your MCP client config — e.g. ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "jobdrop": {
      "command": "jobdrop-mcp-server"
    }
  }
}

That's it — the client launches jobdrop-mcp-server as a stdio subprocess on demand. No daemon, no port, no nix.

Note: prefer the uv tool install path over uvx --from "jobdrop[mcp]" jobdrop-mcp-server in MCP configs. The direct-binary pattern matches what the working reference MCP servers (filesystem, git, etc.) use, and avoids subtle stdio-handoff issues that can occur with the uvx wrapper.

Usage

from jobdrop import scrape_jobs

jobs = scrape_jobs(
    site_name=["insight_global", "clearance_jobs", "kforce", "greenhouse",
               "linkedin", "indeed", "google"],
    search_term="site reliability engineer",
    location="Atlanta, GA",
    results_wanted=20,
    hours_old=720,          # 30-day freshness cap
    country_indeed="usa",
)
print(f"Found {len(jobs)} jobs")
print(jobs[["site", "title", "company", "location", "min_amount", "max_amount", "job_url"]].head())

Parameters

scrape_jobs(
  site_name              list[str] | str — any of the 17 sources above (default: all)
  search_term            str        — keyword query
  google_search_term     str        — Google Jobs override (only filter for `google`)
  location               str        — "City, ST" or ZIP. Each scraper geocodes its own way.
  distance               int        — radius miles, default 50
  is_remote              bool       — remote-only filter (where supported)
  job_type               str        — "fulltime" | "parttime" | "contract" | "internship"
  easy_apply             bool       — direct-board apply only (LinkedIn easy-apply is broken)
  results_wanted         int        — per-site target
  offset                 int        — pagination offset
  hours_old              int        — drop postings older than N hours
  country_indeed         str        — Indeed/Glassdoor country (see list below)
  description_format     str        — "markdown" | "html"
  enforce_annual_salary  bool       — convert hourly/monthly to yearly
  linkedin_fetch_description  bool  — full JD + direct URL (slower)
  linkedin_company_ids   list[int]  — filter LinkedIn by company IDs
  proxies                list[str]  — round-robin proxies, "user:pass@host:port"
  ca_cert                str        — CA cert path for proxies
  user_agent             str        — override the default UA
  verbose                int        — 0 errors / 1 warnings / 2 all
)

Per-scraper limitations

• Indeed — only one of hours_old / (job_type+is_remote) / easy_apply per call.
• LinkedIn — only one of hours_old / easy_apply per call.
• ClearanceJobs — location/remote filters require facet IDs from the dropdown endpoints (not implemented). Filter client-side or scope by keyword.
• InsightGlobal — does not expose client-company name (it's the staffing firm). is_remote is not available in their data.
• Greenhouse — Google indexes some postings after they're filled. Stale 404s are filtered out; the freshness cutoff filters "live but ancient" postings (default 90 days, override with hours_old).

JobPost schema

JobPost
├── id, title, company_name, company_url, job_url
├── location { country, city, state }
├── description
├── is_remote
├── date_posted
├── job_type        fulltime | parttime | contract | internship
├── compensation
│   ├── interval   yearly | monthly | weekly | daily | hourly
│   ├── min_amount, max_amount, currency
│   └── salary_source
├── job_level                                  (LinkedIn, ClearanceJobs)
├── company_industry                           (LinkedIn, Indeed, Greenhouse, Kforce)
├── company_country, company_addresses,
│   company_employees_label, company_revenue_label,
│   company_description, company_logo          (Indeed)
├── skills, experience_range,
│   company_rating, company_reviews_count,
│   vacancy_count, work_from_home_type         (Naukri)
└── emails

Indeed / Glassdoor country list

Pass country_indeed (use the exact name; * = also supported on Glassdoor):

Argentina	Australia*	Austria*	Bahrain
Belgium*	Brazil*	Canada*	Chile
China	Colombia	Costa Rica	Czech Republic
Denmark	Ecuador	Egypt	Finland
France*	Germany*	Greece	Hong Kong*
Hungary	India*	Indonesia	Ireland*
Israel	Italy*	Japan	Kuwait
Luxembourg	Malaysia	Mexico*	Morocco
Netherlands*	New Zealand*	Nigeria	Norway
Oman	Pakistan	Panama	Peru
Philippines	Poland	Portugal	Qatar
Romania	Saudi Arabia	Singapore*	South Africa
South Korea	Spain*	Sweden	Switzerland*
Taiwan	Thailand	Turkey	Ukraine
United Arab Emirates	UK*	USA*	Uruguay
Venezuela	Vietnam*

LinkedIn searches globally and uses only location. ZipRecruiter is US/Canada and uses only location. Bayt searches internationally with only search_term.

Notes

• Most boards cap a single search at ~1000 results.
• LinkedIn rate-limits aggressively around the 10th page of pagination on a single IP. Use proxies.
• For Indeed search-term tuning: it searches the description too. Use -foo to exclude, "exact phrase" for exact match. Example:

  search_term='"site reliability engineer" (kubernetes OR terraform) -recruiter'
  

• For Google: copy the exact filter syntax from a real Google Jobs search and pass it as google_search_term.
• For Greenhouse: keyword + location are passed straight to a Google site:greenhouse.io query, so Boolean operators and quotes work. Don't quote the full "City, ST" — quote the city alone, leave the state bare.

License

MIT. See LICENSE.