LinkedIn Jobs Server

Self-hosted MCP server that scrapes LinkedIn jobs with your authenticated session cookies. Runs in Docker. No Apify, no per-run costs.

Setup

1. Get your cookies

1. Install the Cookie-Editor Chrome extension
2. Log into LinkedIn (use a burner account)
3. Click Cookie-Editor → Export (JSON)
4. Find and copy two values:
   • li_at — your session cookie
   • JSESSIONID — used as CSRF token (value looks like ajax:1234...)

2. Configure

cp .env.example .env
# Edit .env and paste your cookie values

3. Build and test

# Build the Docker image
docker build -t linkedin-mcp .

# Quick smoke test — should print the MCP server startup log
docker run --rm --env-file .env linkedin-mcp
# Ctrl+C to stop

4. Wire into Claude Desktop

Edit your Claude Desktop config file:

• Mac: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "linkedin-jobs": {
      "command": "docker",
      "args": [
        "run", "--rm", "-i",
        "--env-file", "/ABSOLUTE/PATH/TO/linkedin-mcp/.env",
        "linkedin-mcp"
      ]
    }
  }
}

⚠️ Use the absolute path to your .env file. ~/ does not expand here.

5. Restart Claude Desktop

After saving the config, fully quit and reopen Claude Desktop. You'll see a 🔧 tools icon in the chat bar — click it to confirm scrape_jobs, get_job_details, check_cookie, and update_cookies are listed.

6. First conversation

You: Check if my LinkedIn cookie is valid
Claude: [calls check_cookie] ✓ Authenticated as John Doe

You: Scrape 20 AI Engineer or ML Engineer jobs posted in the last 3 days in the US
Claude: [calls scrape_jobs] ...returns full job list with descriptions

---

Cookie refresh (every 30–60 days)

When cookies expire, re-export from Cookie-Editor and tell Claude:

Update my LinkedIn cookies: li_at is "new_value" and jsessionid is "new_value"

Claude will call update_cookies — no container restart needed.

---

Development

# Run with live source reloading
docker compose up

# Inspect MCP tools without Claude Desktop
npx @modelcontextprotocol/inspector docker run --rm -i --env-file .env linkedin-mcp

Project structure

linkedin-mcp/
├── src/
│   ├── server.py        # MCP server — tool definitions and handlers
│   ├── scraper.py       # LinkedIn Voyager API calls (httpx)
│   ├── models.py        # Pydantic models for Job data
│   └── __init__.py
├── Dockerfile
├── docker-compose.yml
├── requirements.txt
├── .env.example
└── .gitignore