LinkedIn MCP

This is a simple implementation of an MCP LinkedIn server using a Selenium and Chromedriver based scraper in the background.

At the moment it only support profile scraping and profile search.

Pre-requisites

Please install uv.

Create a virtual environment and sync the libraries.

uv venv
uv sync

Then install Chromedriver: https://developer.chrome.com/docs/chromedriver/get-started

The whole parsing of the LinkedIn webpage is based on Chromedriver.

Environment

Make sure you have all of these variables in your .env file:

LINKEDIN_USER_1=<user>
LINKEDIN_PASSWORD_1=<pass>

LINKEDIN_USER_2=<user>
LINKEDIN_PASSWORD_2=<pass>

COOKIE_DIR=/var/linkedin/cookies

MCP_HOST=0.0.0.0
MCP_PORT=8050
# Either sse or stdio or streamable-http
MCP_TRANSPORT=streamable-http
MCP_TIMEOUT=150

You can use multiple LinkedIn username and password combinations.

Run in dev mode with MCP Inspector

mcp dev ./src/linkedin_mcp/server/server.py

Run with stdio

python ./src/linkedin_mcp/client/client_stdio.py

Run the server

uv run ./src/linkedin_mcp/server/server.py

Note: If you're experiencing issues with code changes not being picked up, clear the Python cache first:

# Windows PowerShell
.\clear_cache.ps1

# Or manually:
Get-ChildItem -Path . -Filter __pycache__ -Recurse -Directory | Remove-Item -Recurse -Force

Then restart the server. Alternatively, use Python's -B flag to skip bytecode generation:

uv run python -B ./src/linkedin_mcp/server/server.py

Integration with Gemini CLI

This is a modified .settings.json of the Gemini CLI with which you can use for connecting and using the MCP:

{
  "general": {
    "preferredEditor": "vscode"
  },
  "security": {
    "auth": {
      "selectedType": "gemini-api-key"
    }
  },
  "ui": {
    "theme": "Default"
  },
  "mcpServers": {
    "linkedin-scraper": {
      "httpUrl": "http://0.0.0.0:8050/mcp/",
      "transport": "streamable-http"
    }
  }
}