linkedin_mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@linkedin_mcp post a LinkedIn update about our new product launch"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
LinkedIn MCP Server
This README is intentionally minimal and focused on run instructions only.
Technical details moved to
README_TECHNICAL.md.Use
README_TECHNICAL.mdfor architecture, API details, schema, tools reference, and implementation notes.
Prerequisites
Python 3.11+
uvPostgreSQL
Proxy credentials (Oxylabs or IPFoxy)
Related MCP server: LinkedIn MCP Server
1) Run Without Docker (local)
# 1. Install Python dependencies
uv sync
# 2. Install Chromium for Patchright
uv run patchright install chromium
# 3. Prepare environment variables
cp .env.example .env
# Edit .env and set DATABASE_URL + proxy settings
# 4. Run DB migrations
uv run alembic upgrade head
# 5. Create a profile (cookie-first recommended)
uv run python scripts/create_profile.py AE Asia/Dubai --li_at="AQEDAQNxyz..."
# 6. Start HTTP MCP server (for test scripts and HTTP clients)
# Server runs on http://127.0.0.1:8765
uv run python -m linkedin_mcp.http_server
# 7. (Alternative) Start stdio MCP server (for Claude Desktop, Cursor, etc.)
# This waits for JSON-RPC on stdin — it will appear to hang in a terminal.
# Use this only when configured as an MCP server in your client app.
uv run linkedin-mcpLocal profile examples
# Oxylabs with state targeting
uv run python scripts/create_profile.py US America/Los_Angeles --state=california --li_at="AQEDAQNxyz..."
# Oxylabs with city targeting
uv run python scripts/create_profile.py DE Europe/Berlin --city=munich --li_at="AQEDAQNxyz..."
# IPFoxy dedicated proxy per profile (requires PROXY_PROVIDER=ipfoxy)
uv run python scripts/create_profile.py US America/New_York \
--ipfoxy-host=gw.ipfoxy.com \
--ipfoxy-port=12345 \
--ipfoxy-username=your_profile_user \
--ipfoxy-password=your_profile_pass \
--li_at="AQEDAQNxyz..."2) Run With Docker
# 1. Prepare environment variables
cp .env.example .env
# Edit .env and set DATABASE_URL + proxy settings
# 2. Build and start services
docker compose up -d --build
# 3. Check server health
curl http://localhost:8765/health
# 4. Follow logs if needed
docker compose logs -f mcp-server
# 5. Stop services
docker compose downNotes:
docker-compose.ymlruns migrations via a dedicatedmigrationservice beforemcp-serverstarts.Browser state is persisted in Docker volume
browser_state.
3) Common Commands
# -----------------------------------------------------------------------------
# Profile setup and auth scripts
# -----------------------------------------------------------------------------
# scripts/create_profile.py - create a profile (cookie-first recommended)
uv run python scripts/create_profile.py AE Asia/Dubai --li_at="AQEDAQNxyz..."
# scripts/set_cookies.py - update li_at for an existing profile
uv run python scripts/set_cookies.py <profile-uuid> "AQEDAQNxyz..."
# scripts/import_cookies.py - import cookies JSON (exported from browser)
uv run python scripts/import_cookies.py <profile-uuid> data/cookies.json
# scripts/clone_fingerprint.py - clone fingerprint to existing profile
pbpaste | uv run python scripts/clone_fingerprint.py <profile-uuid>
# scripts/extract_fingerprint.js - run in Chrome DevTools Console to copy fingerprint JSON
# 1) Open any page in Chrome
# 2) Open DevTools -> Console
# 3) Paste content of scripts/extract_fingerprint.js and run it
# -----------------------------------------------------------------------------
# Utility scripts (direct browser/session tests; no HTTP server required)
# -----------------------------------------------------------------------------
# scripts/test_fingerprint.py - verify proxy IP + fingerprint attributes
uv run python scripts/test_fingerprint.py <profile-uuid>
# scripts/test_login.py - direct login flow with optional OTP code
uv run python scripts/test_login.py <profile-uuid> [totp_code]
# scripts/test_linkedin_auth.py - validate LinkedIn auth/session state
uv run python scripts/test_linkedin_auth.py <profile-uuid>
# scripts/test_bot_detection.py - inspect bot-detection signals and feed behavior
uv run python scripts/test_bot_detection.py <profile-uuid>
# -----------------------------------------------------------------------------
# HTTP client script
# -----------------------------------------------------------------------------
# scripts/mcp_client.py - basic client connectivity + list tools test
# (requires HTTP server running at localhost:8765)
uv run python scripts/mcp_client.py
# -----------------------------------------------------------------------------
# MCP API test scripts (HTTP server must be running)
# -----------------------------------------------------------------------------
# Start server first (terminal 1)
uv run python -m linkedin_mcp.http_server
# scripts/test_mcp_login.py - login via MCP using stored credentials
uv run python scripts/test_mcp_login.py <profile-uuid>
# scripts/test_open_manual_browser.py - open pinned manual browser session
uv run python scripts/test_open_manual_browser.py <profile-uuid>
# scripts/test_save_session_cookies.py - save cookies from active manual session
uv run python scripts/test_save_session_cookies.py <profile-uuid>
# scripts/test_close_session.py - close active session and persist cookies
uv run python scripts/test_close_session.py <profile-uuid>
# scripts/test_read_feed.py - fetch feed items
uv run python scripts/test_read_feed.py <profile-uuid>
# scripts/test_read_messages.py - fetch inbox/conversations
uv run python scripts/test_read_messages.py <profile-uuid>
# scripts/test_get_profile.py - fetch profile by LinkedIn URL
uv run python scripts/test_get_profile.py <profile-uuid> https://linkedin.com/in/someone/
# scripts/test_get_company.py - fetch company by LinkedIn URL
uv run python scripts/test_get_company.py <profile-uuid> https://linkedin.com/company/microsoft/
# scripts/test_search_people.py - search people by keywords
uv run python scripts/test_search_people.py <profile-uuid> "Software Engineer" US
# scripts/test_search_posts.py - search posts by keywords
uv run python scripts/test_search_posts.py <profile-uuid> ai marketing
# scripts/test_get_post.py - fetch a specific post by URL
uv run python scripts/test_get_post.py <profile-uuid> https://linkedin.com/feed/update/urn:li:activity:123456789
# scripts/test_like_post.py - like a post
uv run python scripts/test_like_post.py <profile-uuid> https://linkedin.com/feed/update/urn:li:activity:123456789
# scripts/test_comment_post.py - comment on a post
uv run python scripts/test_comment_post.py <profile-uuid> https://linkedin.com/feed/update/urn:li:activity:123456789 "Great post!"
# scripts/test_like_and_comment_post.py - like and comment in one call
uv run python scripts/test_like_and_comment_post.py <profile-uuid> https://linkedin.com/feed/update/urn:li:activity:123456789 "Great post!"
# scripts/test_send_connection.py - send connection request with a note
uv run python scripts/test_send_connection.py <profile-uuid> https://linkedin.com/in/someone/ "Hi!"
# scripts/test_send_message.py - send direct message
uv run python scripts/test_send_message.py <profile-uuid> https://linkedin.com/in/someone/ "Hello!"
# scripts/test_send_inbox_message.py - send message by conversation thread or profile compose route
uv run python scripts/test_send_inbox_message.py <profile-uuid> --conversation-id <conversation-id> --message "Hello!"
uv run python scripts/test_send_inbox_message.py <profile-uuid> --participant-profile-url https://linkedin.com/in/someone/ --message "Hello!"
# scripts/test_create_post.py - create personal post with optional image
uv run python scripts/test_create_post.py <profile-uuid> masterlabs_post.txt cost_of_not_automating_follow_up.jpg
# scripts/test_create_company_post.py - create company page post
uv run python scripts/test_create_company_post.py <profile-uuid> https://linkedin.com/company/masterlabs-ai/ masterlabs_post.txt cost_of_not_automating_follow_up.jpg4) Unit Tests
Unit tests use mocks and an in-memory SQLite database — no real browser, network, or PostgreSQL required.
# Run all tests
uv run pytest
# Run with verbose output
uv run pytest -v
# Run a specific test module
uv run pytest tests/test_proxy.py
# Run a specific test class
uv run pytest tests/test_proxy.py::TestApifyProxy
# Run with coverage summary (requires pytest-cov)
uv run pytest --cov=linkedin_mcp --cov-report=term-missingTest coverage
Module | Tests | Description |
| 10 | Fernet encrypt/decrypt round-trip, edge cases, invalid keys |
| 8 | Settings loading from env, singleton, logging setup |
| 30 | Apify/Oxylabs/IPFoxy providers, session IDs, geo-targeting |
| 22 | Stealth launch args, init script generation, fingerprint patches |
| 8 | Navigation retry on transient network errors |
| 20 | Human-like typing, mouse movement, delays, Bezier paths |
| 25 | Profile/Fingerprint/CookieData/CookieStore serialization & persistence |
| 31 | Profile/Cookie/ChallengeEvent/BrowserState CRUD with in-memory SQLite |
| 18 | Tool routing, auth recovery, fatal session error detection |
| 10 | Challenge signal classification, lock response builder |
Technical Documentation
Full technical reference:
README_TECHNICAL.md
Available Tools
21 toolsclose_sessionB
Close a browser session and save cookies.
Inputs: profile_id (required).
comment_postC
Comment on a LinkedIn post.
Inputs: post_url (required), profile_id (required), comment_text (required).
create_company_postA
Create a post on behalf of a company page (requires admin access).
Inputs: content (required), profile_id (required), company_url (required), image_path.
create_postA
Create a personal post on LinkedIn.
Inputs: content (required), profile_id (required), image_path.
get_companyB
Get LinkedIn company details by URL.
Inputs: profile_id (required), company_url (required).
get_postA
Get a LinkedIn post by URL. Returns post details including whether the current account has liked or commented on it.
Inputs: post_url (required), profile_id (required).
get_profileB
Get LinkedIn profile details by URL.
Inputs: profile_id (required), linkedin_url (required), max_posts, include_activity.
get_session_statusC
Get status of a browser session.
Inputs: profile_id (required).
like_and_comment_postA
Like and comment on a LinkedIn post in one operation (single page load).
Inputs: post_url (required), profile_id (required), comment_text (required).
like_postB
Like a LinkedIn post.
Inputs: post_url (required), profile_id (required).
loginA
Login to LinkedIn with credentials and optional 2FA. First tries stored cookies.
Inputs: profile_id (required), email, password, totp_code.
open_manual_browserA
Open and pin a persistent browser window for manual profile actions.
Inputs: profile_id (required), url.
read_feedC
Read LinkedIn feed posts.
Inputs: profile_id (required), max_posts, scroll_count.
read_messagesC
Read LinkedIn messages/conversations.
Inputs: profile_id (required), max_conversations, max_messages_per_conversation.
save_session_cookiesB
Save cookies from active browser session to profile storage.
Inputs: profile_id (required).
search_peopleB
Search for people on LinkedIn.
Inputs: keywords (required), profile_id (required), page, country, +2 more.
search_postsB
Search for LinkedIn posts by keywords.
Inputs: keywords (required), profile_id (required), max_posts, scroll_count.
send_connection_requestC
Send a connection request to a LinkedIn user.
Inputs: profile_id (required), recipient_url (required), note.
send_inbox_messageB
Send a message using conversation-id thread routing or profile-url compose routing.
Inputs: message (required), profile_id (required), conversation_id, existing_thread_only, +1 more.
send_messageC
Send a direct message to a LinkedIn connection.
Inputs: message (required), profile_id (required), recipient_url (required).
set_cookiesA
Set LinkedIn authentication cookies (li_at) for a profile. Use this for purchased accounts.
Inputs: li_at (required), profile_id (required), cookies, jsessionid.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 21 tools
Most tools map to distinct resources/actions, but send_message and send_inbox_message overlap in purpose, and like_and_comment_post partially overlaps with like_post and comment_post. Otherwise clear.
All tools use a consistent verb_noun snake_case pattern (e.g., get_profile, send_message), with only 'login' deviating slightly. Naming is predictable and clear.
At 21 tools, the server is on the heavier side but justifiable given the breadth of LinkedIn actions covered (auth, feed, search, messaging, posting). Each tool addresses a distinct workflow, so it avoids bloat.
Core LinkedIn operations are covered: auth/session management, feed reading, profile/company retrieval, people/posts search, post interaction, messaging, and connection requests. Minor gaps like connection request responses or post editing prevent a perfect score.
Maintenance
Related MCP Connectors
Managed LinkedIn MCP server for AI agents: search, connect, message and enrich on accounts you own.
Hosted LinkedIn MCP server on your own account: 35 tools for search, profiles, inbox, posts, pages.
A hosted LinkedIn MCP server for your own account. Paste one address into Claude, ChatGPT, Cursor or n8n to read your inbox, send messages and invitations, and search people. Free to start. Not made by LinkedIn.
Sales MCP server for AI agents: find, enrich, score and reach leads on LinkedIn, track replies.
Related MCP Servers
- AlicenseAqualityDmaintenanceAn MCP server for LinkedIn automation that enables users to search for jobs, retrieve profile details, manage connections, and read or send messages. It leverages Playwright and Browserbase to interact with LinkedIn through an existing authenticated browser session.61MIT
- AlicenseAqualityDmaintenanceFully featured MCP server that provides automation tools for LinkedIn, supporting browser-based scraping and API-based operations for content management, media uploads, and reactions.64MIT
- AlicenseNot gradedqualityDmaintenanceMCP server for programmable LinkedIn automation via Playwright, offering 20 tools for profile management, messaging, feed interaction, and job searching through real browser automation.18 npmMIT
- AlicenseBqualityDmaintenancePlaywright-powered MCP server for LinkedIn that automates jobs, profile edits, messaging, network actions, and feed posts using a real logged-in browser session.36Apache 2.0