LinkedIn MCP Server

A Model Context Protocol (MCP) server that provides LinkedIn automation tools via Playwright browser control. Designed to work with MCP clients like Claude Desktop over the Stdio transport.

Includes built-in stealth measures to reduce automation detection and a global rate limiter (10-30s randomized delay between actions) for safe, human-paced operation. All tool responses use semantic data reduction — null/empty fields are stripped, text is capped, and only the relevant DOM section is scraped — keeping token usage low.

Quick Start

Prerequisites

• Node.js >= 18 — nodejs.org
• Git

Step 1 — Run the installer (one command)

curl -fsSL https://raw.githubusercontent.com/thomascox/LinkedInMCP/main/install.sh | bash

This single command:

1. Clones the repo to ~/.linkedin-mcp-server/
2. Installs all dependencies and compiles TypeScript
3. Installs the Playwright Chromium browser
4. Writes the linkedin entry into your Claude Desktop config automatically

Step 2 — Restart Claude Desktop

Quit and reopen Claude Desktop. The LinkedIn tools will appear.

Step 3 — Authenticate

In Claude, say: "Use manage_auth_session with action 'capture' to log in to LinkedIn."

A browser window opens. Log in normally. Once you reach the feed, the session is saved and the browser closes.

That's it.

---

Manual setup (if you prefer)

If you'd rather clone to a custom location:

git clone https://github.com/thomascox/LinkedInMCP.git
cd LinkedInMCP
npm install              # also builds dist/ automatically
npx playwright install chromium

Then open ~/Library/Application Support/Claude/claude_desktop_config.json and add:

{
  "mcpServers": {
    "linkedin": {
      "command": "node",
      "args": ["/full/path/to/LinkedInMCP/dist/index.js"]
    }
  }
}

Restart Claude Desktop.

Configuration

The server stores browser data and session state in ~/.linkedin-mcp/:

Path	Purpose
~/.linkedin-mcp/browser-data/	Persistent Chromium user profile
~/.linkedin-mcp/storageState.json	Saved cookies and session data
~/.linkedin-mcp/screenshots/	Easy Apply step screenshots

No manual configuration is required. These directories are created automatically on first use.

Stealth and Safety

Anti-detection

Every browser context launched by the server has stealth patches applied automatically:

• Removes navigator.webdriver flag
• Injects window.chrome runtime stubs
• Populates navigator.plugins with realistic entries
• Overrides WebGL vendor/renderer to mask headless GPU signatures
• Sets window.outerWidth/outerHeight to match inner dimensions
• Patches Permissions.prototype.query for notifications
• Launches Chromium with --disable-blink-features=AutomationControlled
• Sets realistic user agent, locale (en-US), and timezone

Rate Limiter

A global rate limiter enforces a randomized delay of 10-30 seconds between every LinkedIn action. This applies to all tool calls that interact with LinkedIn. The delay is measured from the completion of the previous action, so natural pauses between LLM reasoning steps count toward the wait.

Token Efficiency

All tools use semantic data reduction to minimize context usage:

• compactJson — strips null/empty fields before returning JSON (no indentation)
• DOM scraping scoped to [role="main"] or the specific modal/section — navigation, ads, and tracking scripts are never included
• Long text fields (job descriptions, post bodies, message bodies) capped at the scraper level
• Tool descriptions act as mini-prompts with retry guidance and edge-case handling built in

---

Tools

Authentication

linkedin_ping

Health check to confirm the server is running.

Parameters: None

manage_auth_session

Manage LinkedIn authentication sessions.

Name	Type	Required	Description
action	"capture" | "verify"	Yes	The action to perform

• capture — Opens a headed browser, navigates to login, waits for you to log in, saves session cookies.
• verify — Headless check that the saved session is still valid (loads feed, checks for nav/page content).

Sessions expire after ~24-48 hours. If any tool returns a session-expired error, re-run with capture.

---

Search

search_linkedin

Search LinkedIn for people or jobs. Returns compact JSON results.

Name	Type	Required	Description
category	"PEOPLE" | "JOBS"	Yes	Type of search to perform
keywords	string	Yes	Search keywords
filters	object	No	Optional filters (see below)

Filter options:

Name	Type	Applies to	Description
location	string	Both	Location name (e.g. "San Francisco Bay Area")
remote	"onsite" | "remote" | "hybrid"	Jobs	Work arrangement filter
experienceLevel	"internship" | "entry" | "associate" | "mid-senior" | "director" | "executive"	Jobs	Experience level filter

People result shape: { name, headline, profileUrl }

Job result shape: { jobId, title, company, location, easyApply }

---

Profile

view_profile

View any LinkedIn profile by URL. Returns structured JSON with name, headline, location, about, experience entries, education entries, and skills (up to 25).

Name	Type	Required	Description
profile_url	string (URL)	Yes	Full LinkedIn profile URL (e.g. https://www.linkedin.com/in/username)

manage_profile

Read or update your own LinkedIn profile sections.

Name	Type	Required	Description
action	"get_profile" | "update_section"	Yes	Action to perform
section	"headline" | "about"	For update_section	Section to edit
text	string	For update_section	New content

• get_profile — Returns your headline, location, about, experience, education, and skills as compact JSON.
• update_section — Opens edit modal, clears text, types new content with human-like delay, saves, and waits for the confirmation toast.

manage_experience

Add or edit a LinkedIn experience entry on your profile.

Name	Type	Required	Description
action	"add" | "edit"	Yes	"add" to create a new entry, "edit" to update an existing one
match_title	string	For edit	Partial title text to match the existing entry (case-insensitive). Use get_profile to see current titles.
title	string	For add	Job title
company	string	No	Company name (autocomplete attempted)
employment_type	string	No	Full-time | Part-time | Self-employed | Freelance | Contract | Internship | Apprenticeship | Seasonal
location	string	No	Location (e.g. San Francisco, CA)
location_type	string	No	On-site | Hybrid | Remote
start_month	number	No	Start month (1–12)
start_year	number	No	Start year (e.g. 2022)
end_month	number	No	End month (1–12) — omit for current roles
end_year	number	No	End year — omit for current roles
is_current	boolean	No	Set true if this is a current/active role
description	string	No	Role description or bullet points

---

Connections

send_connection_request

Send a LinkedIn connection request. Returns status: sent | already_connected | pending | limit_reached | connect_button_not_found.

Important: If status is limit_reached, stop — LinkedIn enforces a weekly invitation cap.

Name	Type	Required	Description
profile_url	string (URL)	Yes	Full LinkedIn profile URL
note	string	No	Personal note to include (max 300 chars)

get_connections

List your LinkedIn connections. Returns up to 20 with name, headline, and profile URL.

Name	Type	Required	Description
search	string	No	Filter connections by name

manage_connection_requests

Manage incoming connection requests.

Name	Type	Required	Description
action	"list_received" | "accept" | "decline"	Yes	Action to perform
profile_url	string (URL)	For accept/decline	Profile URL of the person to respond to

• list_received — Returns pending invites with name, headline, profileUrl, and mutualConnections.
• accept / decline — Responds to the matching invitation.

---

Jobs

get_job_details

Get full details for a LinkedIn job posting.

Name	Type	Required	Description
job_id	string	Yes	LinkedIn Job ID (from search_linkedin)

Returns: title, company, location, workplaceType, postedDate, applicantCount, easyApply, description (capped at 2000 chars).

save_job

Bookmark a job for later. Returns status: saved | already_saved.

Name	Type	Required	Description
job_id	string	Yes	LinkedIn Job ID

get_saved_jobs

Retrieve your saved/bookmarked jobs. Returns up to 20 in the same shape as search_linkedin job results.

Parameters: None

---

Easy Apply

start_application

Start an Easy Apply application for a LinkedIn job. Returns the initial form state (fields + summary) and keeps the browser alive for subsequent fill_application_step calls.

Name	Type	Required	Description
job_id	string	Yes	LinkedIn Job ID

If the Easy Apply button is not found, the job may not support Easy Apply or the listing was removed.

fill_application_step

Fill the current Easy Apply form step and advance.

Name	Type	Required	Description
answers	array	Yes	Array of { label, value } pairs matching form field labels
action	"next" | "review" | "submit"	Yes	Step action to take

Auto-unchecks "Follow company" at the review step. Returns validation errors for LLM retry when required fields are missing.

cancel_application

Cancel an in-progress Easy Apply session and close the browser.

Parameters: None

---

Messaging

get_messages

Retrieve the last 10 conversation threads from the LinkedIn messaging inbox. Returns sender name, last message snippet, and a conversationUrl for use with get_conversation.

Parameters: None

get_conversation

Read the full message history of a conversation thread. Returns up to 20 messages with sender, text (capped at 500 chars), and timestamp.

Name	Type	Required	Description
conversation_url	string (URL)	Yes	LinkedIn conversation thread URL (from get_messages)

send_linkedin_message

Send a direct message to a LinkedIn user. Only works for 1st-degree connections.

Name	Type	Required	Description
profile_url	string (URL)	Yes	Full LinkedIn profile URL
message_body	string	Yes	The message text to send

get_unread_count

Get the count of unread messages and notifications from the LinkedIn nav badges.

Parameters: None

Returns: { unreadMessages, unreadNotifications }

---

Feed & Content

get_feed

Fetch the top 10 non-promoted posts from your LinkedIn home feed. Returns author, headline, post text snippet (300 chars), reaction count, comment count, and post URL.

Parameters: None

create_post

Create a new LinkedIn post.

Important: Always confirm post content with the user before calling — posts are immediately visible to your network.

Name	Type	Required	Description
text	string	Yes	The post content to publish
visibility	"anyone" | "connections"	No	Who can see the post (default: "anyone")

react_to_post

React to a LinkedIn post with a specific reaction type.

Name	Type	Required	Description
post_url	string (URL)	Yes	Full LinkedIn post URL (from get_feed)
reaction	"like" | "celebrate" | "support" | "funny" | "love" | "insightful"	Yes	Reaction to apply

If the reaction picker doesn't open, fall back to "like".

---

Notifications

get_notifications

Fetch your 10 most recent LinkedIn notifications. Returns type (connection | job | reaction | comment | mention | birthday | work_anniversary | profile_view | post_share), actor name, text snippet (200 chars), and timestamp.

Parameters: None

---

Project Structure

LinkedInMCP/
├── src/
│   ├── index.ts            # Server entry point, all tool registration
│   ├── config.ts           # Central configuration (paths, server metadata)
│   ├── logger.ts           # Stderr logging utility (prevents JSON-RPC corruption)
│   ├── browser.ts          # Shared browser launch + session + stealth + rate limit
│   ├── stealth.ts          # Anti-detection init scripts (navigator, WebGL, chrome, etc.)
│   ├── rate-limiter.ts     # Global 10-30s randomized delay between actions
│   ├── response-utils.ts   # compactJson, fieldsToSummary, randomDelay
│   └── tools/
│       ├── auth.ts         # manage_auth_session
│       ├── search.ts       # search_linkedin
│       ├── profile.ts      # view_profile, manage_profile
│       ├── experience.ts   # manage_experience
│       ├── connections.ts  # send_connection_request, get_connections, manage_connection_requests
│       ├── jobs.ts         # get_job_details, save_job, get_saved_jobs
│       ├── messaging.ts    # get_messages, get_conversation, send_linkedin_message, get_unread_count
│       ├── feed.ts         # get_feed, create_post, react_to_post
│       ├── notifications.ts# get_notifications, get_unread_count
│       └── easy-apply.ts   # start_application, fill_application_step, cancel_application
├── dist/                   # Compiled JavaScript (generated by npm run build)
├── install.sh              # One-command installer
├── package.json
└── tsconfig.json

Development

# Run from source (TypeScript)
npm start

# Run from compiled output
npm run start:built

# Type-check without emitting
npx tsc --noEmit

# Build to dist/
npm run build

License

ISC