linkedin-mcp-analyzer

MCP server for automated LinkedIn saved jobs analysis with EROI scoring and KB write-back.

Architecture

LinkedIn (saved jobs) → Patchright browser → Scraper → EROI scorer → KB writer
                         6 dimensions: domain 35%, tech 25%, role 20%,
                         growth 10%, formal 5%, location 5%
                         → agregovany_report.md + metadata_stacku.json + git commit

Setup

uv sync
linkedin-mcp --login   # one-time LinkedIn auth
linkedin-mcp --status  # verify session

Usage

# Start MCP server
linkedin-mcp

# Or via MCP client
uv run python -m linkedin_mcp_custom

MCP Tools

Tool	Description
get_saved_jobs	List saved jobs from LinkedIn tracker
get_job_details	Scrape full posting for a job ID
analyze_saved_jobs	Full pipeline: scrape → EROI → KB write-back

EROI Scoring

Dimensions

Dimension	Weight	What it measures
Domain	35%	Industrial automation vs adjacent vs noise
Tech	25%	Skill overlap (content-aware match ratio × coverage)
Role	20%	Engineering role vs "fake engineer" (service/sales)
Growth	10%	Strategic employer (Siemens, Google…) vs growth vs other
Formal	5%	Degree requirements with flexibility detection
Location	5%	Remote/hybrid/CZ vs distant/office-only

Thresholds

Score	Verdict
≥65	SLEDOVAT
50–64	MEDIUM
40–49	HRANICNI
<40	NESLEDOVAT

Special patterns

• Fake engineer: title says "Engineer" but content is service/sales
• Positioning match: strong role match compensates for domain gap
• Degree flexibility: "equivalent practical experience" adds ~5%
• Electronics manufacturing: SMT/PCBA keywords cap domain score

Tests

uv run python tests/test_eroi_regression.py   # 6 regression tests
uv run python tests/test_kb_writer.py          # 4 KB writer tests

Phases

Phase	What	Tag
0	Project scaffold	v0.1.0
1	Browser + auth (Patchright)	—
2	Scraping engine (LinkedInExtractor)	—
3	MCP tools (get_saved_jobs, analyze…)	v0.2.0
4	EROI engine (6 scorers)	v0.3.0
5	KB writer (report + metadata + git commit)	v0.4.0
6	DevOps (ruff, pre-commit, README)	v0.5.0