linkedin-jobs-scraper

Scrape public available jobs on Linkedin using headless browser.

• 🔑 Session management: long running session with auto recovery
• 🐢 Adaptive rate limiting: auto adjust scraping speed to avoid rate limiting
• 📄 Fields parser: job_id, link, apply_link, title, company, company_link, company_employee_count, company_img_link, place, description, description_html, date, date_text, insights, salary, is_easy_apply, applicant_count, benefits, reposted
• 🔍 Filters: relevance, time, type, experience, industry, salary, remote, company
• 📡 Events hooks: data, metrics, errors
• 🚀 Headless support: can run in background
• ⌨️ Command line interface: scrape straight from your shell, no code required
• 🐍 Programmatic API: drive it from Python with full control

[!WARNING] For personal or educational use only. All extracted data is publicly available on LinkedIn and remains owned by LinkedIn. I am not responsible for any inappropriate use of data extracted through this library.

Sponsored by

NinjaPear

Scrape+Enrich rich B2B profile data in real-time.

Table of Contents

• Requirements
• Installation
• Usage
  • CLI
  • Programmatic
  • Pinning a location by geoId
• Authentication
• Adaptive Rate limiting
• Filters
• Company filter
• Logging
• License

Requirements

• Chrome or Chromium
• Python >= 3.10

Selenium automatically downloads a Chromedriver matching your Chrome version. You can also specify custom paths like so:

scraper = LinkedinScraper(
    chrome_executable_path='/path/to/chromedriver',
    chrome_binary_location='/path/to/chrome',
)

Installation

Install package:

pip install linkedin-jobs-scraper

Usage

Both a command line interface and a Python API are supported. Before your first scrape you need to authenticate once. The quickest way is the CLI:

lijs login --chrome-user-data-dir ~/.linkedin-jobs-scraper

See Authentication for all the options (Chrome profile, cookie pair, headless machines).

CLI

Installing the package also installs a command line interface. It mirrors the programmatic API but scrapes a single query per invocation (there is no max_workers or chrome_options on the CLI).

Two equivalent commands are installed:

linkedin-jobs-scraper --help   # full command
lijs --help                    # short alias

The package is also runnable as a module:

python -m linkedin_jobs_scraper --help

Subcommands

jobs

Search jobs matching the provided query, locations and filters:

lijs jobs "software engineer" --location "United States" --location "Remote" --limit 50 --chrome-user-data-dir <path>
lijs jobs "data scientist" --geo-id 103644278 --time week --type full-time,contract \
  --experience mid-senior --workplace remote --salary 120k --apply-link --chrome-user-data-dir <path>

• Positional query — the search keywords.
• --location NAME (repeatable) or --geo-id ID (repeatable) — mutually exclusive; a geoId pins the search deterministically (see Pinning a location by geoId).
• --limit N — maximum jobs to scrape, 0 for unlimited (default 25).
• --apply-link — resolve the external apply link for each job (slower).
• --skip-promoted-jobs — skip promoted jobs.
• --page-offset N — number of result pages to skip (default 0).

Filters use kebab-case values. Single-valued: --relevance {relevant,recent}, --time {any,day,week,month}, --salary {40k,60k,80k,100k,120k,140k,160k,180k,200k}, --company-jobs-url URL. Repeatable or comma-separated: --type (full-time, part-time, temporary, contract, internship, volunteer, other), --experience (internship, entry-level, associate, mid-senior, director, executive), --workplace (on-site, remote, hybrid), --industry (e.g. software-development, banking, it-services), --job-function (e.g. engineering, sales, information-technology), --benefits (e.g. medical, vision, dental), --commitments (e.g. work-life-balance, social-impact). Boolean toggles: --easy-apply (only LinkedIn Easy Apply jobs), --under-10-applicants (only jobs with fewer than 10 applicants). Run lijs jobs --help for the full list of values.

job

Lookup a single job id or a /jobs/view/<id> url, with an optional --apply-link:

lijs job 3690634839 --chrome-user-data-dir <path>
lijs job https://www.linkedin.com/jobs/view/3690634839 --apply-link --chrome-user-data-dir <path>

login

linkedin-jobs-scraper login --chrome-user-data-dir ~/.linkedin-jobs-scraper

Opens a visible browser to sign in once into a reusable Chrome profile, then prints the cookie pair ready to export. Requires --chrome-user-data-dir; also accepts --chrome-executable-path and --chrome-binary-location.

Driver flags

Shared by jobs and job: --no-headless, --slow-mo SECONDS, --no-adaptive-slow-mo, --page-load-timeout SECONDS, --chrome-executable-path PATH, --chrome-binary-location PATH, --chrome-user-data-dir DIR, --interactive-login.

Output

DATA is written to stdout; progress, metrics and errors go to stderr, so piping the data stream stays clean.

• -f, --out-format {table,jsonl,json,csv} — output format.
• -o, --out-path PATH — destination; - means stdout.
• --fields a,b,c — comma-separated list of fields to emit.
• --all-fields — emit every available field.
• --vertical — render one field per line.
• --raw — emit the raw record unformatted.

When --out-format is omitted the format is inferred from the --out-path extension (.csv, .json, .jsonl); with no path it defaults to table on a TTY and jsonl when piped or written to a file.

In the table format on a TTY, URL fields are rendered as clickable terminal hyperlinks (OSC 8): columns show a compact label (host and last path segment) while clicking opens the full URL. Structured formats (jsonl/json/csv) always carry the full, unmodified URLs.

In the table format on a TTY, cell values are colour-coded per column (title, company, place, date, link) to make rows easier to scan; --no-color (or the NO_COLOR environment variable) disables it.

Global flags

--quiet, -v/-vv (repeatable, increases verbosity), --no-color, --version. --no-color is accepted on every subcommand, before or after it.

Exit codes

Programmatic

import logging
from linkedin_jobs_scraper import LinkedinScraper
from linkedin_jobs_scraper.events import Events, EventData, EventMetrics, EventBegin
from linkedin_jobs_scraper.query import Query, QueryOptions, QueryFilters, Location
from linkedin_jobs_scraper.filters import RelevanceFilters, TimeFilters, TypeFilters, ExperienceLevelFilters, \
    OnSiteOrRemoteFilters, SalaryBaseFilters

# Change root logger level (default is WARN)
logging.basicConfig(level=logging.INFO)


# Fired once per query/location before scraping starts, carrying LinkedIn's approximate
# total result count (job_total is -1 when it could not be parsed)
def on_begin(data: EventBegin):
    print('[ON_BEGIN]', data.job_total)


# Fired once for each successfully processed job
def on_data(data: EventData):
    print('[ON_DATA]', data.title, data.company, data.company_link, data.date, data.date_text, data.link, data.insights,
          len(data.description))


# Fired once for each page (25 jobs)
def on_metrics(metrics: EventMetrics):
    print('[ON_METRICS]', str(metrics))


def on_error(error):
    print('[ON_ERROR]', error)


def on_end():
    print('[ON_END]')


scraper = LinkedinScraper(
    chrome_executable_path=None,  # Custom Chrome executable path (e.g. /foo/bar/bin/chromedriver)
    chrome_binary_location=None,  # Custom path to Chrome/Chromium binary (e.g. /foo/bar/chrome-mac/Chromium.app/Contents/MacOS/Chromium)
    chrome_options=None,  # Custom Chrome options here
    headless=True,  # Overrides headless mode only if chrome_options is None
    max_workers=1,  # How many threads will be spawned to run queries concurrently (one Chrome driver for each thread)
    slow_mo=0.8,  # Minimum seconds slept between jobs, to avoid 'Too many requests 429' errors. Minimum 0.2, default 0.8
    adaptive_slow_mo=True,  # Slow down automatically when Linkedin throttles the run, then ease back. See 'Rate limiting'
    page_load_timeout=40,  # Page load timeout (in seconds)
    chrome_user_data_dir=None,  # Chrome profile reused across runs, so the scraper keeps its own session. See 'Authentication'
    interactive_login=False  # Sign in by hand on the first run, when chrome_user_data_dir holds no session. Requires a display
)

# Add event listeners
scraper.on(Events.BEGIN, on_begin)
scraper.on(Events.DATA, on_data)
scraper.on(Events.ERROR, on_error)
scraper.on(Events.END, on_end)

queries = [
    Query(
        options=QueryOptions(
            limit=27  # Limit the number of jobs to scrape. Use 0 to scrape all available jobs (LinkedIn serves up to ~1000).
        )
    ),
    Query(
        query='Engineer',
        options=QueryOptions(
            locations=['Europe', Location(geo_id='103644278', name='United States')],  # Plain name, or pin a geo by geoId. See 'Pinning a location by geoId'
            apply_link=True,  # Try to extract apply link (easy applies are skipped). If set to True, scraping is slower because an additional page must be navigated. Default to False.
            skip_promoted_jobs=True,  # Skip promoted jobs. Default to False.
            page_offset=2,  # How many pages to skip
            limit=5,
            filters=QueryFilters(
                company_jobs_url='https://www.linkedin.com/jobs/search/?f_C=1441%2C17876832%2C791962%2C2374003%2C18950635%2C16140%2C10440912&geoId=92000000',  # Filter by companies.                
                relevance=RelevanceFilters.RECENT,
                time=TimeFilters.MONTH,
                type=[TypeFilters.FULL_TIME, TypeFilters.INTERNSHIP],
                on_site_or_remote=[OnSiteOrRemoteFilters.REMOTE],
                experience=[ExperienceLevelFilters.MID_SENIOR],
                base_salary=SalaryBaseFilters.SALARY_100K
            )
        )
    ),
]

scraper.run(queries)

Additional EventData fields

Alongside the core fields, each EventData also carries:

• salary: pay range when LinkedIn shows one (from the fit-level insight or the salary rail card), otherwise ''.
• is_easy_apply: True when the listing uses LinkedIn Easy Apply, False for an external apply flow.
• applicant_count: the applicant segment of the top card (e.g. '27 applicants'), otherwise ''.
• benefits: list of featured benefit labels, empty when none are shown.
• reposted: True when the listing was reposted (derived from the date text).

date is an ISO YYYY-MM-DD string. It comes from the card's exact <time datetime> when available, otherwise it is approximated from the relative date text (weeks, months and years approximated at 7, 30 and 365 days), and is '' only when no date text could be parsed.

Scraping a single job

When you already know the job you want, scrape_job fetches it directly by url or id, bypassing search and pagination. It accepts a bare numeric id or a full /jobs/view/<id> url, emits a single Events.DATA event on success, and Events.ERROR (without raising) on failure. A dead or expired id, one that points to a job that no longer exists, emits Events.NOT_FOUND (carrying an EventNotFound with the job_id) rather than Events.ERROR, since a missing job is not a scraping error.

from linkedin_jobs_scraper import LinkedinScraper
from linkedin_jobs_scraper.events import Events, EventData, EventNotFound


def on_data(data: EventData):
    print('[ON_DATA]', data.title, data.company, data.company_link, data.date_text, data.link,
          data.insights, len(data.description))


def on_error(error):
    print('[ON_ERROR]', error)


def on_not_found(data: EventNotFound):
    print('[ON_NOT_FOUND]', data.job_id)


scraper = LinkedinScraper(
    headless=True,
    max_workers=1,
    slow_mo=0.8,
)

scraper.on(Events.DATA, on_data)
scraper.on(Events.ERROR, on_error)
scraper.on(Events.NOT_FOUND, on_not_found)

# By bare id
scraper.scrape_job('4455383771')

# Or by full url
scraper.scrape_job('https://www.linkedin.com/jobs/view/4455383771/')

# Pass apply_link=True to also extract the external apply link (slower)
scraper.scrape_job('4455383771', apply_link=True)

The single-job path reads every field from the job detail panel, so a few card-only fields are not populated: company_img_link and the promoted flag. date has no exact <time datetime> here, so it is approximated from date_text. query, location are empty and job_index is -1, as there is no search context.

Pinning a location by geoId

A location entry can be a plain string (a place name LinkedIn resolves for you) or a Location that pins the geo deterministically through LinkedIn's own geoId:

from linkedin_jobs_scraper.query import Location

Location(geo_id='103644278', name='United States')

geo_id is what pins the search: it is sent as the geoId URL param, and the location=<name> param is omitted entirely (LinkedIn lets geoId win over a name in a conflict). name is only a human label — it is used for logs and set on EventData.location — so it can be anything, or left out (the geo_id is then used as the label).

To find a real geoId, run the search on LinkedIn in a browser, then read the geoId= value from the resolved URL.

Authentication

The scraper needs a LinkedIn session. The recommended and tested way is a Chrome profile on a local machine (see below). The cookie-based modes are supported alternatives, but LinkedIn may refuse them in some environments (for example CI or a server), so they are not guaranteed everywhere. All modes keep the session alive on their own, so a long run is not interrupted when LinkedIn expires it.

1. Chrome profile

Sign in once into a Chrome profile that the scraper then reuses:

linkedin-jobs-scraper login --chrome-user-data-dir ~/.linkedin-jobs-scraper

A browser window opens on the sign in page. Sign in there, ticking "Keep me logged in" — that is what makes the profile reusable. The password is typed into the browser: nothing in this package reads, stores or transmits it.

Then point the scraper at the same profile:

scraper = LinkedinScraper(chrome_user_data_dir='~/.linkedin-jobs-scraper')

To skip the separate command and have the first run do the sign in instead:

scraper = LinkedinScraper(
    chrome_user_data_dir='~/.linkedin-jobs-scraper',
    interactive_login=True,
)

interactive_login requires chrome_user_data_dir and waits up to 10 minutes for a human, so leave it off (the default) anywhere nobody is watching, such as CI or a server. Chrome locks a profile directory, so max_workers is forced to 1 whenever chrome_user_data_dir is set.

2. Cookie pair

You can use LinkedIn's remember me cookies (li_rm and bcookie) as environment variables to obtain a session, an option when a browser window is not available (you still need a machine with a browser window to obtain them the first time). Both variables are required. LinkedIn may refuse this mode in some environments (for example CI), so it is not guaranteed everywhere; prefer the Chrome profile when you can.

export LI_RM_COOKIE='<li_rm value>'
export LI_BCOOKIE='<bcookie value>' # keep the double quote " characters the value contains
python your_app.py

Get the two values by running the sign in command above on a machine that has a display: it prints them at the end, quoted and ready to export.

[!WARNING] Do not copy these two cookies out of your browser's developer tools: use the sign in command described above instead.

Setting chrome_user_data_dir as well is worth it if the host has storage that survives across runs: the session is then reused instead of being requested again at the start of each run.

Fallback: a bare session cookie

LI_AT_COOKIE takes the li_at session cookie on its own. This one can be copied straight out of your own Chrome browser. Sign in, then open Chrome developer tools:

Go to tab Application, then from the left panel select Storage -> Cookies -> https://www.linkedin.com, locate the row named li_at and copy the Value column.

LI_AT_COOKIE=<your li_at cookie value here> python your_app.py

This cookie cannot be renewed: LinkedIn expires it after a while, and a run that loses it stops. Expect to replace it by hand. As with the cookie pair, LinkedIn may refuse it in some environments; prefer the Chrome profile when you can.

Begin event

BEGIN fires once per query/location, before any job is scraped, carrying an EventBegin with LinkedIn's approximate total result count for that search. job_total is -1 when the count could not be parsed. Combined with limit=0 (scrape all available jobs, LinkedIn serves up to ~1000), it lets a caller know upfront roughly how many results a query has:

from linkedin_jobs_scraper.events import Events, EventBegin

def on_begin(data: EventBegin):
    print('total results reported by LinkedIn:', data.job_total)

scraper.on(Events.BEGIN, on_begin)

Session events

SESSION_REFRESHED fires whenever the scraper ends up holding a session cookie different from the one it was given. Listen to it if you have nowhere else to store a session and want to reuse it on the next run:

from linkedin_jobs_scraper.events import Events, EventSession

def on_session_refreshed(session: EventSession):
    print('store this for the next run:', session.li_at)

scraper.on(Events.SESSION_REFRESHED, on_session_refreshed)

INVALID_SESSION fires when every credential supplied was refused, immediately before the run aborts with InvalidCookieException. It takes no arguments:

def on_invalid_session():
    print('LinkedIn refused every credential')

scraper.on(Events.INVALID_SESSION, on_invalid_session)

[!NOTE] Changed in 6.0.0: INVALID_SESSION used to fire whenever a session cookie went missing, which normally happened right before a new one was issued and the run carried on. It now fires only when authentication has actually failed. If you were using it to know when to harvest a fresh cookie, use SESSION_REFRESHED instead.

Not found event

NOT_FOUND (scraper:not-found) fires when a single-job scrape (scrape_job) targets a job that does not exist or is no longer available. It carries an EventNotFound with the job_id that was requested. Throttling and page-load failures stay a silent skip, since neither says anything about whether the job exists:

from linkedin_jobs_scraper.events import Events, EventNotFound

def on_not_found(data: EventNotFound):
    print('job no longer available:', data.job_id)

scraper.on(Events.NOT_FOUND, on_not_found)

Adaptive Rate limiting

Requests failing with the status code 429 mean you are sending too many requests and Linkedin is throttling them. Two parameters control this:

• slow_mo: seconds slept between jobs. Higher is safer, at least 0.2, default 0.8.
• max_workers: how many queries run concurrently. One worker is recommended.

slow_mo sets the fastest the run will ever go, not a fixed delay: with adaptive_slow_mo on (the default) the run starts at that speed and slows itself down whenever Linkedin pushes back.

• On every 429, the delay between jobs doubles, up to min(10, slow_mo * 10) seconds.
• After 20 jobs in a row without a 429, the delay shrinks back towards slow_mo, and never goes below it.
• When a whole page is throttled, the scraper waits and asks for it again: first 5s, then 15s, then 45s. A short burst of throttling no longer ends the query.

Pass adaptive_slow_mo=False to make slow_mo a fixed delay instead.

The METRICS event reports both numbers:

• throttled: how many 429s the run has met.
• pace: the delay currently slept between jobs.

Filters

It is possible to customize queries with the following filters:

• RELEVANCE:
  • RELEVANT
  • RECENT
• TIME:
  • DAY
  • WEEK
  • MONTH
  • ANY
• TYPE:
  • FULL_TIME
  • PART_TIME
  • TEMPORARY
  • CONTRACT
• EXPERIENCE LEVEL:
  • INTERNSHIP
  • ENTRY_LEVEL
  • ASSOCIATE
  • MID_SENIOR
  • DIRECTOR
• ON SITE OR REMOTE:
  • ON_SITE
  • REMOTE
  • HYBRID
• INDUSTRY:
  • AIRLINES_AVIATION
  • BANKING
  • CIVIL_ENGINEERING
  • COMPUTER_GAMES
  • ENVIRONMENTAL_SERVICES
  • ELECTRONIC_MANUFACTURING
  • FINANCIAL_SERVICES
  • INFORMATION_SERVICES
  • INVESTMENT_BANKING
  • INVESTMENT_MANAGEMENT
  • IT_SERVICES
  • LEGAL_SERVICES
  • MOTOR_VEHICLES
  • OIL_GAS
  • SOFTWARE_DEVELOPMENT
  • STAFFING_RECRUITING
  • TECHNOLOGY_INTERNET
• BASE SALARY:
  • SALARY_40K
  • SALARY_60K
  • SALARY_80K
  • SALARY_100K
  • SALARY_120K
  • SALARY_140K
  • SALARY_160K
  • SALARY_180K
  • SALARY_200K
• JOB FUNCTION:
  • ACCOUNTING_AUDITING
  • ADMINISTRATIVE
  • ADVERTISING
  • BUSINESS_DEVELOPMENT
  • CONSULTING
  • DISTRIBUTION
  • DESIGN
  • EDUCATION
  • ENGINEERING
  • FINANCE
  • GENERAL_BUSINESS
  • HEALTH_CARE_PROVIDER
  • HUMAN_RESOURCES
  • INFORMATION_TECHNOLOGY
  • LEGAL
  • MANAGEMENT
  • MANUFACTURING
  • MARKETING
  • OTHER
  • PUBLIC_RELATIONS
  • PRODUCT_MANAGEMENT
  • PROJECT_MANAGEMENT
  • QUALITY_ASSURANCE
  • RESEARCH
  • SALES
  • SUPPLY_CHAIN
  • TRAINING
• BENEFITS:
  • MEDICAL
  • VISION
  • DENTAL
  • RETIREMENT_401K
  • PENSION_PLAN
  • PAID_MATERNITY_LEAVE
  • PAID_PATERNITY_LEAVE
  • COMMUTER_BENEFITS
  • STUDENT_LOAN_ASSISTANCE
  • TUITION_ASSISTANCE
  • DISABILITY_INSURANCE
• COMMITMENTS:
  • DIVERSITY_EQUITY_INCLUSION
  • ENVIRONMENTAL_SUSTAINABILITY
  • WORK_LIFE_BALANCE
  • SOCIAL_IMPACT
  • CAREER_GROWTH_AND_LEARNING
• EASY APPLY: easy_apply=True restricts results to jobs with LinkedIn Easy Apply.
• UNDER 10 APPLICANTS: under_10_applicants=True restricts results to jobs with fewer than 10 applicants.
• COMPANY:
  • See below

See the following example for more details:

from linkedin_jobs_scraper.query import Query, QueryOptions, QueryFilters
from linkedin_jobs_scraper.filters import RelevanceFilters, TimeFilters, TypeFilters, ExperienceLevelFilters, \
    OnSiteOrRemoteFilters, IndustryFilters, SalaryBaseFilters, JobFunctionFilters, BenefitsFilters, CommitmentsFilters
query = Query(
    query='Engineer',
    options=QueryOptions(
        locations=['United States'],        
        apply_link=True,
        skip_promoted_jobs=True,
        limit=5,
        filters=QueryFilters(
            relevance=RelevanceFilters.RECENT,
            time=TimeFilters.MONTH,
            type=[TypeFilters.FULL_TIME, TypeFilters.INTERNSHIP],
            experience=[ExperienceLevelFilters.INTERNSHIP, ExperienceLevelFilters.MID_SENIOR],
            on_site_or_remote=[OnSiteOrRemoteFilters.REMOTE],
            industry=[IndustryFilters.IT_SERVICES],
            base_salary=SalaryBaseFilters.SALARY_100K,
            job_function=[JobFunctionFilters.ENGINEERING],
            benefits=[BenefitsFilters.MEDICAL, BenefitsFilters.VISION],
            commitments=[CommitmentsFilters.WORK_LIFE_BALANCE],
            easy_apply=True,
            under_10_applicants=True
        )
    )
)

Industry Filter

You will probably need to add the industry filter to the IndustryFilters class in filters.py

To find the numeric code for the industry:

1. Perform the search on LinkedIn in a browser, with the industry filter applied.
2. The numeric code is in the URL, immediately after f_I . For example URL https://www.linkedin.com/jobs/search/?currentJobId=3661007408&distance=25&f_E=3%2C4&f_I=43%2C46%2C41%2C45&f_JT=F%2CC&geoId=102257491&keywords=Product%20Owner&refresh=true contains text f_I=43%2C46%2C41%2C45 indicating a filter is applied on industry codes 43, 46, 41 and 45.

Company Filter

It is also possible to filter by company using the public company jobs url on LinkedIn. To find this url you have to:

1. Login to LinkedIn using an account of your choice.
2. Go to the LinkedIn page of the company you are interested in (e.g. https://www.linkedin.com/company/google).
3. Click on jobs from the left menu.

4. Scroll down and locate See all jobs or See jobs button.

5. Right click and copy link address (or navigate the link and copy it from the address bar).
6. Paste the link address in code as follows:

query = Query(    
    options=QueryOptions(        
        filters=QueryFilters(
            # Paste link below
            company_jobs_url='https://www.linkedin.com/jobs/search/?f_C=1441%2C17876832%2C791962%2C2374003%2C18950635%2C16140%2C10440912&geoId=92000000',        
        )
    )
)

Logging

Package logger can be retrieved using namespace li:scraper. Default level is INFO. It is possible to change logger level using environment variable LOG_LEVEL or in code:

import logging

# Change root logger level (default is WARN)
logging.basicConfig(level = logging.DEBUG)

# Change package logger level
logging.getLogger('li:scraper').setLevel(logging.DEBUG)

# Optional: change level to other loggers
logging.getLogger('urllib3').setLevel(logging.WARN)
logging.getLogger('selenium').setLevel(logging.WARN)

License

MIT License

If you like the project and want to contribute you can donate something here!