Rayyan9477 linkedin mcp

An MCP server that gives AI assistants full access to LinkedIn — search jobs, view profiles and companies, generate AI-powered resumes and cover letters, and track applications. Built with the official MCP Python SDK (FastMCP).

# Core installation
pip install -e .

# With AI features (resume/cover letter generation, profile analysis)
pip install -e ".[ai]"

# With PDF export
pip install -e ".[pdf]"

# Everything
pip install -e ".[all]"

cp .env.example .env

Edit .env with your credentials:

LINKEDIN_USERNAME=your_email@example.com
LINKEDIN_PASSWORD=your_password
ANTHROPIC_API_KEY=sk-ant-...    # Optional — enables AI features

Standalone:

linkedin-mcp

With Claude Desktop — add to your claude_desktop_config.json:

{
  "mcpServers": {
    "linkedin": {
      "command": "linkedin-mcp"
    }
  }
}

With Claude Code — add to .mcp.json:

{
  "linkedin": {
    "command": "linkedin-mcp"
  }
}

Templates: modern · professional · minimal (resume) | professional · concise (cover letter) Formats: html · md · pdf

Status values: interested · applied · interviewing · offered · rejected · withdrawn

src/linkedin_mcp/
├── server.py                    # FastMCP entry point — 13 tools, 1 resource
├── config.py                    # Settings from .env (frozen dataclass)
├── exceptions.py                # 7-class exception hierarchy
├── models/
│   ├── linkedin.py              # Profile, Job, Company models (Pydantic v2)
│   ├── resume.py                # Resume & cover letter content models
│   └── tracking.py              # Application tracking model
├── services/
│   ├── linkedin_client.py       # LinkedIn API wrapper (async via asyncio.to_thread)
│   ├── job_search.py            # Job search with TTL caching
│   ├── profile.py               # Profile/company access with caching
│   ├── resume_generator.py      # AI-enhanced resume generation
│   ├── cover_letter_generator.py
│   ├── application_tracker.py   # Local JSON-based application tracking
│   ├── cache.py                 # Unified JSON file cache with TTL
│   ├── template_manager.py      # Jinja2 sandboxed template engine
│   └── format_converter.py      # HTML → PDF/Markdown conversion
├── ai/
│   ├── base.py                  # Abstract AI provider interface
│   └── claude_provider.py       # Anthropic Claude implementation
└── templates/
    ├── resume/                  # modern.j2, professional.j2, minimal.j2
    └── cover_letter/            # professional.j2, concise.j2

• Official MCP SDK — Uses FastMCP with @mcp.tool() decorators, not a custom protocol implementation
• Async throughout — All sync LinkedIn API calls wrapped in asyncio.to_thread() to avoid blocking
• Layered architecture — Tools → Services → Client, with caching at the service layer
• AI is optional — Core LinkedIn features work without an Anthropic API key; AI enhances resume/cover letter generation
• Security hardened — Jinja2 SandboxedEnvironment, WeasyPrint SSRF protection, path traversal guards, credential redaction, input validation

All settings are loaded from environment variables (.env file supported):

# Install with all dependencies
pip install -e ".[all,dev]"

# Run tests (82 tests)
pytest

# Run with coverage
pytest --cov=linkedin_mcp

# Lint
ruff check src/ tests/

Tests cover all layers: config, models, services (cache, tracker, job search, profile, resume/cover letter generation, LinkedIn client formatters, format converter), AI provider, and MCP tool handlers.

Once connected, ask your AI assistant:

"Search for remote Python developer jobs in the US"

"Show me the profile for satyanadella"

"Generate a resume from my LinkedIn profile tailored to job 3847291056"

"Create a cover letter for job 3847291056 using the concise template"

"Track my application for the Senior Engineer role at Google — status: applied"

"List all my applications that are in the interviewing stage"

"Analyze my LinkedIn profile and suggest improvements"

MIT