Why AI Job Hunting Tools Fail (And How We Fixed It with a 4-Tier Dynamic ATS Engine)

If you've experimented with AI-powered job search scripts or LLM job matching agents, you've probably noticed two glaring problems:

1. Dead Listings & Ghost Jobs: Web-scraped search results constantly return postings that closed weeks ago or passed their deadline.
2. Naive Keyword Matching: A mechanical engineer or industrial product manager gets matched to crypto trading or forex brokers simply because both job descriptions happened to mention "Agile", "Jira", or "KPIs".

To solve this, I built JobsDirect-MCP — an open-source, candidate-agnostic Model Context Protocol (MCP) server that connects Claude Desktop, Cursor, and AI agents directly to global and regional job markets.

What is JobsDirect?

JobsDirect runs completely locally and privately on your machine using the Model Context Protocol standard. It aggregates live listings across 6 active portals:

• LinkedIn (Global & local)
• Himalayas (Curated remote tech with transparent salary ranges)
• Kariyer.net (Turkey's top job board)
• Remote.co (Curated remote roles)
• CareerHound.io (Hidden & direct company postings)
• HiredChina (Bilingual & international roles)

1. Automated Freshness & Dead-Listing Gating

Instead of blindly parsing search results, JobsDirect implements strict freshness and closure detection:

• Closure Keyword Recognition: Automatically identifies explicit closure announcements ("no longer accepting applications", "ilan yayından kaldırılmıştır", "başvuruya kapalıdır", "this position has been filled").
• Deadline Comparison: Extracts application deadline dates dynamically and compares them against datetime.today().
• Search Freshness Timelimits: Constrains web queries to live index windows (7d to 14d) so search engines don't return stale cached pages.

2. The 4-Tier Dynamic Weighted ATS Engine

Rather than relying on primitive keyword counts, JobsDirect evaluates candidates through 4 distinct weighted criteria:

1. Tier 1: Industry & Domain Affinity (40% Weight): Dynamically builds an industry taxonomy vector from the candidate's career footprint vs. the employer's domain. If domain overlap is 0%, domain affinity drops to 0/40, preventing false-positive cross-industry matches.
2. Tier 2: Minimum Qualifications Gate (20% Weight): Verifies required degree hierarchy (B.Sc, M.Sc, PhD) and minimum Years of Experience (YoE).
3. Tier 3: Core Functional & Technical Skills (30% Weight): Calculates exact technical stack and methodology intersection.
4. Tier 4: Logistics & Language Gate (10% Weight + Hard Gate): Flags security clearance barriers, work authorization restrictions, and language fluency requirements.

3. Executive Word (.docx) Export

Tailoring a CV or cover letter shouldn't require copy-pasting markdown blocks into Word. JobsDirect includes a dedicated export_to_docx tool that renders:

• Tailored CVs utilizing the STAR framework (Situation, Task, Action, Result).
• Targeted 3-Act Cover Letters.
• Formatted application question answers.

All styled with professional typography (Calibri/Calibri Light, executive navy headers, bullet spacing) and saved directly to your desktop.

How to Set It Up in 2 Minutes

1. Clone & Install

git clone https://github.com/netizen4-bit/jobs-direct.git
cd jobs-direct
pip install -r requirements.txt

2. Configure Claude Desktop

Add JobsDirect to your claude_desktop_config.json:

{
  "mcpServers": {
    "jobs-direct": {
      "command": "python",
      "args": ["C:/path/to/jobs-direct/jobs_direct.py"]
    }
  }
}

Now open Claude and type:

"Find remote operations manager jobs posted within the last 7 days, evaluate fit against my resume, and generate tailored application docs on my Desktop."

Open Source & Community

JobsDirect is completely open-source under the MIT license.

⭐ Check out the code, star the repo, or contribute on GitHub:
👉 https://github.com/netizen4-bit/jobs-direct