Job Search MCP

A configuration-driven Model Context Protocol (MCP) server for searching jobs directly from company career websites.

Allows AI assistants (Claude, ChatGPT) to search and extract job listings from 5 pharmaceutical companies with extensible architecture for unlimited company additions.

Current Coverage: Amgen, Bayer, GSK, Novartis, Pfizer
Expandable to: 250+ companies (healthcare, tech, finance sectors)

Features

• ✅ MCP-compatible job search tool for AI assistants
• ✅ Configuration-driven company support (JSON-based, no code changes needed)
• ✅ Extract detailed job information: title, description, requirements, expiry date, apply links
• ✅ Intelligent error tracking: classify 404s, timeouts, network errors, parse errors
• ✅ Generate CSV reports for batch processing
• ✅ Zero external parsing dependencies (pure regex-based extraction)
• ✅ TypeScript strict mode with full type safety
• ✅ Company-specific HTML parsers (Workday, Eightfold AI, Drupal platforms)
• ✅ Optional web demo at /demo for manual job search
• ✅ Comprehensive test suite (4 professional tests)

Technology Stack

• Protocol: Model Context Protocol (MCP) SDK (TypeScript)
• Runtime: Node.js v18+ (LTS recommended: v18, v20, v22)
• Language: TypeScript 5.3+ (strict mode)
• Parsing: Regex-based HTML extraction (no Puppeteer, jsdom, or Cheerio)
• Build: TypeScript Compiler (tsc)
• Optional Web UI: React + Next.js + Tailwind CSS (for manual searching)
• Testing: Native Node.js (no jest/mocha required)

Quick Start

As an MCP Server (For AI Integration)

1. Install & Build:

npm install
npm run build

2. Start the MCP Server:

npm start
# Server runs on stdio (ready for Claude Desktop, Cursor, or other MCP clients)

3. Configure in Claude Desktop (~/.claude/claude_desktop_config.json):

{
  "mcpServers": {
    "job-search": {
      "command": "node",
      "args": ["path/to/dist/server.js"]
    }
  }
}

4. Use in Claude:

User: "Find me senior manager jobs at Amgen and Pfizer"
Claude: (uses MCP search tool)
Claude: "I found 12 senior manager positions with details..."

As a Local CLI Tool

# Install dependencies
npm install

# Build TypeScript
npm run build

# Run job search tests
npm run test
node dist/test/test-manager-jobs.js

Optional: Web Demo for Manual Browsing

Deploy the included React demo to Vercel (optional):

# Deploy demo at https://[your-app].vercel.app/
vercel deploy

Project Architecture

As MCP Server (Primary)

Claude / AI Assistant
        ↓
    MCP Client Protocol (stdio)
        ↓
   MCP Server (src/server.ts)
        ↓
  Search Tool Handler
        ↓
┌──────────────────────────────────────┐
│  SearchExecutor (src/search-executor.ts)
│  - Orchestrates job searches
│  - Fetches from career site URLs
│  - Parses HTML for job listings
└──────────────────────────────────────┘
        ↓
┌──────────┬──────────┬──────────┬──────────┬──────────┐
│          │          │          │          │          │
Amgen    Bayer      GSK     Novartis    Pfizer
│          │          │          │          │
└──────────┴──────────┴──────────┴──────────┴──────────┘
        ↓
ExtractorRegistry (src/extractors/)
- 5 Company-specific parsers
- Extract: jobTitle, description, requirements, applyLink
- Track errors with classification
        ↓
Return JSON to AI Assistant

Optional: Web Demo

User → Web Browser
        ↓
  React Component (app/demo/page.tsx)
        ↓
  Next.js API Route (app/api/search-jobs/route.ts)
        ↓
  SearchExecutor (same as MCP uses)
        ↓
  Results + CSV reports

Configuration

src/config.json is the source of truth for the companies that the project supports.

Current: 5 companies (Amgen, Bayer, GSK, Novartis, Pfizer)
Expandable: Add unlimited companies via JSON configuration (no code changes needed)

Example:

{
  "projectname": "Job Search MCP",
  "sites": [
    {
      "name": "Amgen",
      "search_url": "https://amgen.wd1.myworkdayjobs.com/Careers?q={SEARCH_TERM}"
    },
    {
      "name": "Bayer",
      "search_url": "https://bayer.eightfold.ai/careers?query={SEARCH_TERM}"
    }
  ]
}

Adding New Companies

To add a new company:

1. Create config entry: Add to src/config.json with company name and search URL
2. Create site definition: Add sites/company-name.json with search parameters
3. Create extractor: Add src/extractors/company-name.ts with HTML parsing rules (only if using new platform)
4. Register extractor: Add to src/extractors/index.ts

No rebuild needed - configuration is loaded at runtime.

Company Platform Support

Platform	Companies	File
Workday	Amgen, Pfizer, GSK	src/extractors/amgen.ts, etc.
Eightfold AI	Bayer	src/extractors/bayer.ts
Drupal	Novartis	src/extractors/novartis.ts

Extensibility: Adding 50+ more companies only requires JSON config + reusable platform extractors "name": "Pfizer", "search_url": "..." } ] }

Only the company name and a usable search URL need to be supplied when adding a new company.

## Site Definitions

Each company is represented by a separate file under `sites/`.

For example:

```text
sites/amgen.json

The structure must follow test/sample.json.

A site definition contains:

• company name
• career URL
• search URL
• supported search parameters
• parameter labels
• parameter types
• available parameter values

The parameter structure is intentionally an array rather than fixed JSON keys because different career websites expose different search parameters.

For example, one site may expose:

location
country
jobType

while another may expose:

location
timeType
LocationCountry
jobFamilyGroup
workerSubType

The MCP must not assume that every company supports the same parameters.

Job Extractors

The project includes site-specific job extractors that parse individual job posting URLs and extract detailed information.

Extracted Data

Each extractor retrieves:

• Job Title - Position name
• Job Description - Full job description/responsibilities (excludes headers/footers)
• Eligibility - Requirements, qualifications, and skills
• Expiry Date - Application closing date (YYYY-MM-DD format, blank if not available)
• Apply Link - Direct URL to apply (may differ from job posting URL)

Available Extractors

src/extractors/
├── types.ts                  # JobExtractor interface & types
├── amgen.ts                  # Amgen (Workday-based)
├── pfizer.ts                 # Pfizer (Workday-based)
├── bayer.ts                  # Bayer (Eightfold AI)
├── gsk.ts                    # GSK (Workday-based)
├── novartis.ts               # Novartis (Drupal)
└── index.ts                  # ExtractorRegistry

Usage Example

import { ExtractorRegistry } from './src/extractors/index.js';

const registry = new ExtractorRegistry();
const amgenExtractor = registry.getExtractor('amgen');

const result = await amgenExtractor?.extract(
  'https://amgen.wd1.myworkdayjobs.com/job/India---Hyderabad/Assoc-Director---Data-Product-Mgmt_R-219150'
);

if (result?.success && result.data) {
  console.log(result.data.jobTitle);
  console.log(result.data.jobDescription);
  console.log(result.data.eligibility);
}

Testing

The project includes a comprehensive test suite for validating search and extraction functionality.

Test Suite Overview

All tests are self-contained TypeScript files that can be run independently:

npm run build
node dist/test/[test-name].js

Available Tests

1. test-config.ts - Configuration Loading Test

Tests that company configurations load correctly from src/config.json.

node dist/test/test-config.js

Purpose: Validates configuration structure and company discovery Output: Lists available companies and their search URLs

---

2. test-search.ts - Job Search Test

Tests the search functionality across all companies.

node dist/test/test-search.js

Purpose: Verifies that searches return valid job URLs Output: Search results for "Manager" jobs from each company Note: Requires internet connectivity to actual career sites

---

3. test-extractors.ts - Job Extraction Test

Tests that job detail extraction works for each company's job URLs.

node dist/test/test-extractors.js

Purpose: Validates job title, description, and eligibility extraction Output: Extraction success rate and field details Note: Requires real job URLs from test-search.ts output

---

4. test-manager-jobs.ts - End-to-End Integration Test

Complete pipeline test: searches for jobs → extracts details → generates reports

node dist/test/test-manager-jobs.js

Purpose: Full integration test with error tracking and CSV report generation Output:

• test/manager-jobs-success.csv - Successfully extracted job data
• test/manager-jobs-errors.csv - Extraction errors (404s, timeouts, etc.)
• Console summary showing success rate and error breakdown

Running All Tests

npm run build
node dist/test/test-config.js
node dist/test/test-search.js
node dist/test/test-extractors.js
node dist/test/test-manager-jobs.js

Test Output Files

Generated CSV reports are stored in the test/ folder:

• manager-jobs-success.csv - Successful job extractions
• manager-jobs-errors.csv - Failed extraction attempts with error codes
• Sample HTML files for debugging

These files are generated during test runs and can be safely deleted. They are in .gitignore.

test/sample.json

test/sample.json defines the expected structure for individual company files.

It is a schema-by-example/template rather than a company registry.

The current example uses parameters such as location, timeType, LocationCountry, jobFamilyGroup, and workerSubType.

BUILD.md

BUILD.md contains instructions for the AI/development process that builds the MCP from the company configurations.

The build process should:

1. Read src/config.json.
2. Process every company listed in sites.
3. Visit/analyze the supplied search URL.
4. Determine the company's actual career/search structure.
5. Discover the available search parameters and their values.
6. Generate or update the corresponding sites/<company>.json.
7. Ensure the generated file follows the structure defined by test/sample.json.
8. Build/update the common MCP implementation.
9. Validate that all configured sites can be searched.

UPDATE.md

See ai/UPDATE.md for instructions for rebuilding the project when a new release is created.

When src/config.json changes, the AI must rebuild all company definitions, not only newly added companies.

This is intentional.

Existing career sites can change their:

• search URLs
• query parameters
• filter names
• filter values
• career-site structure
• ATS implementation

Therefore, every release should re-check existing sites/*.json files against the current live career sites.

src/config.json updated
       │
       ▼
Rebuild ALL sites
       │
       ├── New company → create site JSON
       │
       └── Existing company → re-analyze and update
       │
       ▼
Rebuild common MCP
       │
       ▼
Validate

Project Structure

JobSearchMCP/
├── src/                      # Source code & configs
│   ├── server.ts             # MCP server entry point
│   ├── search-executor.ts    # Search execution & parsing
│   ├── config-loader.ts      # Configuration loader
│   ├── types.ts              # TypeScript types
│   ├── config.json           # Company registry
│   ├── site_configurations.json
│   └── site_analysis.json
├── sites/                    # Company-specific configs
│   ├── amgen.json
│   ├── pfizer.json
│   ├── novartis.json
│   ├── bayer.json
│   └── gsk.json
├── test/                     # Tests & test data
│   ├── test-*.js             # Test scripts
│   ├── sample.json           # Configuration template
│   └── *.html                # Sample HTML files
├── ai/                       # AI development notes (Gitignored)
│   ├── AI.md
│   └── UPDATE.md
├── reports/                  # Documentation
│   ├── IMPLEMENTATION.md
│   ├── ANALYSIS_GUIDE.md
│   ├── MCP_USAGE.md
│   └── MIGRATION.md
├── dist/                     # Compiled JavaScript
├── package.json              # Dependencies & scripts
├── tsconfig.json             # TypeScript config
└── README.md                 # This file

Technology

Runtime: Node.js Language: TypeScript MCP SDK: Official Model Context Protocol TypeScript SDK Configuration: JSON

Design Principle

The project separates site-specific knowledge from common MCP logic.

sites/*.json
    = How a particular company career site works

MCP implementation
    = How to search any configured company

AI
    = Understand the user's request and select/use the appropriate
      company search configuration

The MCP should not contain hard-coded assumptions about parameters such as location, remote, full_time, or job_type.

A parameter only exists for a company if that company's career site actually supports it or exposes the information required by the configuration.

Goal

The goal is to create a reusable job-search MCP where adding companies is primarily a matter of adding their search URLs to config.json, allowing the AI build process to discover and maintain the site-specific configurations automatically.