LinkedIn MCP Server

A Model Context Protocol (MCP) server for LinkedIn. Search people, companies, and jobs, scrape profiles, and retrieve structured JSON data from any MCP-compatible AI client.

https://github.com/user-attachments/assets/50cd8629-41ee-4261-9538-40dc7d30294e

Built with FastMCP, Patchright, and a clean hexagonal architecture.

Features

Person Profile Sections

The get_person_profile tool supports granular section scraping. Request only the sections you need:

• Main profile (always included) — name, headline, location, followers, connections, about, profile image

• Experience — title, company, dates, duration, description, company logo

• Education — school, degree, dates, description, school logo

• Contact info — email, phone, websites, birthday, LinkedIn URL

• Interests — people, companies, and groups followed

• Honors and awards — title, issuer, description

• Languages — language name and proficiency level

• Posts — recent activity with reactions and timestamps

• Recommendations — received and given, with author details

Company Profile Sections

• About (always included) — overview, website, industry, size, headquarters, specialties, logo

• Posts — recent feed posts with engagement metrics

• Jobs — current open positions

Job Search Filters

The search_jobs tool supports the following filters:

Applying for Jobs

The apply_for_job tool automates the "Easy Apply" process for a given job. It will intelligently cycle through the application steps and submit the application. If it encounters a required custom question it cannot answer, it will securely close the dialog and notify the client that manual completion is required.

Sharing Posts & Content Generation

The share_post tool publishes text directly to your LinkedIn feed. Standard MCP architecture dictates that content generation should remain in your LLM client (e.g. Claude Desktop).

To generate viral, high-engagement posts, simply prompt Claude with your desired template. For example:

"Claude, generate a high engagement LinkedIn post about how I failed my first MCP server project. Use the 'LoRA + API synthetic pairs' viral template and include relevant hashtags. Once you generate it, use the share_post tool to publish it."

Error Handling

All tools utilize a robust, structured JSON error schema. If an error occurs (e.g., session expired, rate limit hit), the server returns a detailed message that MCP clients like Claude can easily interpret:

{
  "error": "RATE_LIMIT_EXCEEDED",
  "message": "[search_jobs] LinkedIn rate limit detected. Please wait ~5 minutes before retrying.",
  "retryable": true,
  "timestamp": "2026-03-21T17:15:00.000000+00:00"
}

Prerequisites

• Python 3.12 or later

• uv package manager

• A LinkedIn account for authentication

Quick Start

1. Clone and install

git clone https://github.com/[Your GitHub Username]/linkedin-mcp-server.git
cd linkedin-mcp-server
uv sync

2. Install browser

This project uses Patchright (a patched fork of Playwright) for browser automation. You need to install the browser binaries before first use:

uv run patchright install

Windows users: If the command above fails with program not found, run instead:

uv run python -m patchright install

3. Authenticate with LinkedIn

uv run linkedin-mcp-server --login

A browser window will open. Log in to LinkedIn and the session will be persisted locally at ~/.linkedin-mcp-server/browser-data.

4. Run the server

stdio transport (default — for Claude Desktop, Cursor, and similar clients):

uv run linkedin-mcp-server

HTTP transport (for remote clients, the MCP Inspector, etc.):

uv run linkedin-mcp-server --transport streamable-http --host 0.0.0.0 --port 8000

Client Integration

Claude Desktop / Cursor

Add to your MCP configuration file:

{
  "mcpServers": {
    "linkedin": {
      "command": "uv",
      "args": [
        "--directory", "/path/to/linkedin-mcp-server",
        "run", "linkedin-mcp-server"
      ]
    }
  }
}

MCP Inspector

npx @modelcontextprotocol/inspector

Then connect to http://localhost:8000/mcp if using HTTP transport.

Configuration

Configuration follows a strict precedence chain: CLI args > environment variables > .env file > defaults.

CLI Arguments

Environment Variables

Create a .env file in the project root:

# Server
LINKEDIN_TRANSPORT=stdio
LINKEDIN_HOST=127.0.0.1
LINKEDIN_PORT=8000
LINKEDIN_LOG_LEVEL=WARNING

# Browser
LINKEDIN_HEADLESS=true
LINKEDIN_SLOW_MO=0
LINKEDIN_TIMEOUT=10000
LINKEDIN_VIEWPORT_WIDTH=1280
LINKEDIN_VIEWPORT_HEIGHT=720
LINKEDIN_CHROME_PATH=
LINKEDIN_USER_AGENT=
LINKEDIN_USER_DATA_DIR=~/.linkedin-mcp-server/browser-data

Architecture

The project follows a hexagonal (ports and adapters) architecture with strict layer separation:

src/linkedin_mcp_server/
├── domain/              # Core business logic — zero external dependencies
│   ├── models/          # Data models (Person, Company, Job, Search)
│   ├── parsers/         # HTML to structured data parsers
│   ├── exceptions.py    # Domain exceptions
│   └── value_objects.py # Immutable configuration and content objects
├── ports/               # Abstract interfaces
│   ├── auth.py          # Authentication port
│   ├── browser.py       # Browser automation port
│   └── config.py        # Configuration port
├── application/         # Use cases — orchestration layer
│   ├── scrape_person.py
│   ├── scrape_company.py
│   ├── scrape_job.py
│   ├── search_people.py
│   ├── search_jobs.py
│   └── manage_session.py
├── adapters/            # Concrete implementations
│   ├── driven/          # Infrastructure adapters (browser, auth, config)
│   └── driving/         # Interface adapters (CLI, MCP tools, serialization)
└── container.py         # Dependency injection composition root

Design Decisions

• Ports and adapters — Domain logic is fully decoupled from infrastructure. The browser engine, MCP framework, and configuration source can all be swapped independently.

• Dependency injection — A single Container class acts as the composition root and is the only place that imports concrete adapter classes.

• Structured JSON output — LinkedIn HTML is parsed into typed Python dataclasses, then serialized to JSON for reliable LLM consumption.

• Session persistence — Browser state is saved to disk, so authentication is required only once.

Development

Setup

uv sync --group dev
uv run pre-commit install

Running tests

uv run pytest

With coverage:

uv run pytest --cov=linkedin_mcp_server

Linting and formatting

This project uses Ruff for both linting and formatting. Pre-commit hooks will run these automatically on each commit.

# Lint
uv run ruff check .

# Lint and auto-fix
uv run ruff check . --fix

# Format
uv run ruff format .

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contributing

Contributions are welcome. Please read the contributing guide for details on the development workflow and submission process.

Disclaimer

This tool is intended for personal and educational use. Scraping LinkedIn may violate their Terms of Service. Use responsibly and at your own risk. The authors are not responsible for any misuse or consequences arising from the use of this software.