Your AI assistant wants your LinkedIn password

What it does

This is a Model Context Protocol (MCP) server that lets AI assistants like Claude interact with LinkedIn through your own account. It exposes roughly twenty tools for scraping profiles, companies, jobs, messages, and your home feed—turning LinkedIn’s web UI into structured data that an LLM can query and act on.

The interesting bit

It runs a real browser. The server uses Patchright (a Playwright fork) to drive Chromium, log in as you, and scrape pages—serializing all tool calls to protect a single shared session. That means it can do anything you can do on LinkedIn.com, but it also means you’re trusting a headless browser with your credentials and your inbox.

Key highlights

• Covers the full LinkedIn surface: people search with connection-degree filters, company employees, job details, message threads, even sidebar recommendations
• Ships as a PyPI package (linkedin-scraper-mcp), a one-click .mcpb bundle for Claude Desktop, and a Docker image
• Persistent browser profile stored locally under ~/.linkedin-mcp/profile/
• Supports both stdio and streamable HTTP transports
• Apache 2.0 licensed, 2,248 stars

Caveats

• Several social features are currently broken: connect_with_person, get_conversation, and send_message all have open issues (#407, #432, #442, etc.)
• First-run setup is finicky: browser downloads happen in the background, early tool calls may error out, and LinkedIn can throw captchas or demand mobile-app confirmation
• Tool calls queue serially; heavy scrapes on cold-start Chromium can hit the default 180-second timeout

Verdict

Worth a look if you want your AI assistant to do LinkedIn legwork—recruiting, competitive research, job tracking—without writing custom scrapers. Skip it if the idea of piping your professional identity and DMs through a third-party browser automation layer makes you twitchy.