linkedin-autoapply

Automated LinkedIn Easy Apply — scrape jobs, score them against your CV with Claude, and submit applications without touching a browser.

A CLI tool that scrapes LinkedIn Easy Apply listings, scores each job against your CV using Claude AI, and submits applications via Playwright. All config lives at ~/.linkedin_autoapply/config.json and is collected once via an interactive setup wizard.

Disclaimer: This tool automates interaction with LinkedIn, which may violate the LinkedIn User Agreement. Use at your own risk. The author accepts no liability for account restrictions, bans, or any other consequences. You are solely responsible for compliance with applicable terms of service and laws.

---

Prerequisites

• Python 3.11+
• A Claude API key (see Getting a Claude API Key)
• A LinkedIn account (Easy Apply is a feature on individual job postings, not an account setting — no setup needed)
• Your CV as a PDF file
• Playwright Chromium — installed automatically during setup

Salary filtering is GBP-only. The min_salary value is compared in British pounds. Non-GBP salaries (USD, EUR, etc.) are stored as text but excluded from numeric filtering. Hourly rates are annualised at ×1,880 and daily rates at ×220. If you target non-UK roles, set min_salary to 0 to disable salary filtering.

Installation

git clone https://github.com/pravindurgani/linkedin-autoapply
cd linkedin-autoapply
python -m venv venv
source venv/bin/activate    # Windows: venv\Scripts\activate
pip install -e .
linkedin-autoapply setup    # required — creates config, installs Playwright browser

Why a virtual environment? Python 3.12+ on macOS and many Linux distros enforces PEP 668 and will refuse pip install outside a venv with error: externally-managed-environment. Always use a venv.

The setup wizard validates your Claude API key, stores your LinkedIn password in the system keychain (not on disk), and installs Playwright Chromium if needed.

Headless Linux note: On headless Linux without a secret service daemon (e.g. a VPS), the password cannot be saved to keyring and will be prompted on each run.

First-Run Walkthrough

============================================================
  linkedin-autoapply — First-Run Setup
============================================================

Step 1/8: Claude API key
  Get yours at: https://console.anthropic.com
  API key (sk-ant-...):
  Validating... OK

Step 2/8: LinkedIn account email
  LinkedIn email: jane@example.com

Step 3/8: LinkedIn password
  LinkedIn password:

Step 4/8: Job titles to search for
  Enter one per line. Empty line when done.
  Title 1 (or Enter to finish): Data Analyst
  Title 2 (or Enter to finish): Analytics Engineer
  Title 3 (or Enter to finish):

Step 5/8: Location
  Location (e.g. London, UK) [London]: London, UK

Step 6/8: Minimum annual salary (GBP)
  Min salary (e.g. 90000): 85000

Step 7/8: CV and contact details
  Path to your CV PDF: /Users/jane/Documents/CV.pdf
  Full name: Jane Smith
  Contact email: jane@example.com
  Phone number (e.g. +447000000000): +44700000000

Step 8/8: Visa & Work Authorisation
  Do you have the right to work in your target country? [y/n]: y
  [1] Citizen / permanent resident
  [2] Pre-settled / settled status
  [3] Valid work visa (no sponsorship needed)
  [4] Other
  Select [1-4]: 1

If you answer n (require sponsorship), you'll see:

  [1] Skilled Worker visa (UK)
  [2] Other visa type
  Select [1-2]: 1

This affects job scoring (non-sponsoring companies are penalised) and form-filling (visa/sponsorship screening questions are answered automatically).

============================================================
  Configuration Summary
============================================================
  Claude API key:  sk-ant-...xxxxxx
  LinkedIn email:  jane@example.com
  Job titles:      Data Analyst, Analytics Engineer
  Location:        London, UK
  Min salary:      £85,000
  CV:              /Users/jane/Documents/CV.pdf
  Name:            Jane Smith
  Contact email:   jane@example.com
  Phone:           +44700000000
  Sponsorship:     Not required (Citizen / permanent resident)
  Password:        stored in system keyring
============================================================

Save this configuration? [y/n]: y

Configuration saved to /Users/jane/.linkedin_autoapply/config.json

Playwright Chromium browser not found.
Install it now? (recommended) [y/n]: y

First Login and 2FA

LinkedIn may trigger two-factor authentication or a CAPTCHA on the first automated login. A headless browser cannot complete these challenges, so the session will fail silently.

Run the scrape command in visible mode once to solve the challenge manually:

linkedin-autoapply scrape --visible

A browser window opens. Complete any verification LinkedIn presents. Once you reach your feed, the session cookie is saved to ~/.linkedin_autoapply/linkedin_cookies.json and all subsequent runs proceed headlessly. If LinkedIn re-challenges after a gap (days/weeks of inactivity), re-run --visible.

Command Reference

Command	Description
linkedin-autoapply run	Full pipeline: scrape → score → CV review → apply → export
linkedin-autoapply run --skip-cv-review	Skip the Claude CV review before applying
linkedin-autoapply run --dry-run	Scrape and score only, no applications submitted
linkedin-autoapply run --visible	Full pipeline with visible browser (for 2FA/CAPTCHA)
linkedin-autoapply run --max-applies 5	Limit applications per run (default: 15)
linkedin-autoapply run --retry-failed	Clear failed application records and re-queue them
linkedin-autoapply setup	Run the setup wizard (required on first install, re-run to update config)
linkedin-autoapply scrape	Scrape LinkedIn jobs only — no scoring or applying
linkedin-autoapply scrape --visible	Scrape with visible browser (use for first login)
linkedin-autoapply status	Print database stats (jobs scraped, scored, applied, above threshold)
linkedin-autoapply export	Export results to ~/.linkedin_autoapply/application_report.csv
linkedin-autoapply --verbose run	Enable debug logging (note: --verbose goes before the command)

scrape vs run: run always scrapes first, then scores, then applies. Running scrape separately is useful if you want to inspect jobs in the database (status) before applying. To score and apply without re-scraping, there is no separate command — run is the full pipeline.

How the Pipeline Works

CV review

Before applying, Claude Sonnet reviews your CV against your target job titles and provides a structured assessment: strengths, weaknesses, ATS keyword gaps, and a readiness score out of 10. You are then asked:

Are you happy to proceed with applications? [y/n]:

If you answer n, the pipeline stops — no applications are submitted. If the API call fails (quota exhausted, auth error), you're prompted to continue without the review. Skip entirely with --skip-cv-review.

Scoring

Each scraped job is scored 0–100 by Claude Haiku based on title match, company, description content, salary, and visa/sponsorship fit. Only jobs scoring 70 or above (the default threshold) enter the apply queue. The threshold is set in auto_apply/config.py (SCORE_THRESHOLD = 70).

Rate limiting

Each run applies to at most --max-applies jobs (default: 15). A random delay of 45–90 seconds is inserted between applications. The tool tracks daily application counts and warns when approaching LinkedIn's suspected restriction thresholds. A hard daily cap of 15 is the default (configurable via max_daily_applications in config.json).

Getting a Claude API Key

1. Go to console.anthropic.com and create an account
2. Navigate to API Keys and generate a new key

The tool uses Claude Haiku for job scoring and form-fill (fast and cheap) and Claude Sonnet for the optional CV review step.

Configuration Storage

All config is stored at ~/.linkedin_autoapply/ (your home directory):

~/.linkedin_autoapply/
├── config.json              # Your settings (chmod 600 — readable only by you)
├── jobs.db                  # SQLite database of scraped/scored/applied jobs
├── cv_text.txt              # Cached CV text extraction
├── linkedin_cookies.json    # Saved LinkedIn session — treat as a password
└── application_report.csv   # Exported results

Security note: linkedin_cookies.json contains a fully-authenticated LinkedIn session. Treat it like a password — do not share, commit, or include in backups without encryption.

Your LinkedIn password is not stored in config.json. It is stored in your system keychain (macOS Keychain / Linux Secret Service / Windows Credential Manager) via the keyring library.

Visa & sponsorship

The wizard sets three keys in config.json based on your Step 8 answers:

"requires_sponsorship": false,
"work_authorisation": "citizen",
"visa_notes": ""

These feed into job scoring (sponsorship-requiring candidates are penalised for non-sponsoring companies) and form-filling (visa/sponsorship screening questions are answered automatically). If these keys are absent (legacy config), the tool prompts at runtime instead.

Customising the title filter

config.json contains two lists you can edit directly to change which job titles are considered:

"title_must_contain": ["data", "analytics", "machine learning", ...],
"title_exclude":      ["intern", "director", "contract", ...]

Set title_must_contain to [] to disable the keyword requirement and include all scraped titles. The defaults are tuned for data/analytics roles — adjust for your field.

Troubleshooting

error: externally-managed-environment during pip install You're installing outside a virtual environment. Python 3.12+ enforces this. Create and activate a venv first (see Installation).

LinkedIn CAPTCHA or 2FA on subsequent runs Session cookies expire after prolonged inactivity. Re-run with --visible to solve the challenge manually:

linkedin-autoapply scrape --visible

Claude API quota exhausted mid-run The pipeline logs the error and stops applying. Top up your API credit balance, then re-run. Already-scored jobs are cached in jobs.db and won't re-consume API calls.

Failed applications Use --retry-failed to clear failed records and re-queue them:

linkedin-autoapply run --retry-failed --skip-cv-review

jobs.db growing large after many runs The database accumulates jobs across runs. To start fresh, delete it:

rm ~/.linkedin_autoapply/jobs.db

The next run creates a new database automatically.

CLI entry point not found If linkedin-autoapply isn't on your PATH (e.g. venv not activated), use:

python -m auto_apply run

Verifying installation Run the test suite to confirm everything is working:

pip install -e ".[dev]"
pytest

Anti-Detection & Human Behaviour Hardening

The tool includes several measures to reduce automated-browsing detection signals:

• Viewport randomisation — each session uses a randomised browser viewport (1260–1420 x 780–900) instead of a fixed resolution, reducing fingerprint consistency.
• Humanised timing — all fixed sleep() delays are replaced with ±30% jitter ranges (random.uniform), producing non-repeating, non-integer intervals between actions.
• Mouse movement simulation — before clicking buttons (Easy Apply, form navigation, autocomplete options), the mouse moves to the element with coordinate jitter and a realistic multi-step trajectory.
• Scroll and reading delay — on job detail pages, the browser scrolls in two steps with pauses proportional to the description length (~200 wpm reading speed, capped at 8 seconds).
• Character-by-character typing — login credentials and short form answers are typed with per-keystroke delays (30–120ms) rather than instant fill(), matching human typing patterns.
• Micro-pauses between form fields — a short random delay (0.3–1.0s) is inserted before each field interaction to simulate reading the question.
• Stealth browser patches — playwright-stealth suppresses common headless-browser fingerprints (navigator.webdriver, missing plugins, HeadlessChrome UA substring).

These measures operate inline with no additional configuration, dependencies beyond playwright-stealth, or config keys.