How to Connect LinkedIn to Claude, ChatGPT and Cursor

A LinkedIn MCP server gives an AI agent (Claude, ChatGPT, Cursor and others) a set of LinkedIn tools it can call through the Model Context Protocol: search people and posts, read who engaged, enrich a profile, comment, connect, send a DM, schedule a post. You talk to your assistant in plain language, and it picks the right tool.

Looking for the product page with the server URL and what Apex MCP can do? See LinkedIn MCP Server. This guide is the how-to.

That part is easy to build. The hard part is everything around it: whose limits apply, how fast the agent is allowed to move, what needs your approval, and how you stop it. This guide covers all of that, then shows you how to connect LeadShark's Apex MCP to your assistant in a few steps.

What is a LinkedIn MCP server?

MCP (Model Context Protocol) is an open standard for connecting AI assistants to outside tools. An MCP server publishes a list of tools, each with a name, a description and the inputs it accepts. The assistant reads that list and decides which tool to call to get your request done.

A LinkedIn MCP server is one where those tools act on LinkedIn. So instead of opening five tabs to find prospects, read their recent posts, check whether you've talked before and write a note, you say:

Find heads of RevOps at B2B SaaS companies in my 2nd-degree network, show me what they've commented on lately, and draft a connection note for the three best fits.

The agent brings the reasoning. The MCP server brings the hands.

Is there an official LinkedIn MCP server?

As of October 2026, LinkedIn doesn't offer an official MCP server for acting from your personal profile. Every "LinkedIn MCP" you'll find is third-party, and they differ mostly in one thing: what they do with your account and how much control you keep over it.

Why a "governed" LinkedIn MCP matters

A lot of open-source LinkedIn MCP projects take your logged-in session and let the agent fire off actions as fast as it can think. That works in a demo. On a real account it's how you get restricted. LinkedIn watches for behavior that doesn't look human: bursts of actions, odd hours, the same thing over and over.

An agent with unlimited LinkedIn access will do exactly that, because nothing tells it not to. So the question to ask any LinkedIn MCP server is not "what can it do?" but "what stops it?"

Here's what governance looks like in LeadShark's Apex MCP:

Your limits, shared with your automations. DMs, comment replies and daily/weekly connection requests all draw from the same caps you set in the dashboard. An agent can't route around them.
The agent can't raise its own budget. Through manage_activity_limits an agent can check what's left, lower a cap or pause a channel immediately. Raising a cap is staged for the account owner to approve.
Pacing. Every write is velocity-paced and honors a shared cooldown, so actions are spread out instead of fired in a burst.
Off-hours. If you've turned on off-hours, the agent won't send DMs, comments, reactions or connection requests between 10 PM and 6 AM in your timezone. Reads still work.
Approve-first safety mode. Put the account in queue mode and every write is staged for your review instead of sent. Nothing reaches LinkedIn until you release it.
Emergency Stop. One switch (in the dashboard under Agent → Logs) blocks every agent write on the account instantly: automations, MCP, everything. Reads keep working until you resume.
Do-not-engage is enforced. Archive a lead and DMs to them are refused at the send step, not just "discouraged" in a prompt.
A full audit log. Every governed action is logged. Your agent can read its own history with list_actions (what it did, what was refused and why, what's pending approval), and you can see the same trail in the dashboard.

There's also a client-side layer. In Claude you can set read-only tools to "Always allow" and leave the write tools on "Ask every time", so Claude checks with you before each comment, connection or message.

None of this makes LinkedIn automation risk-free. It does mean the agent operates inside rules you set, at a pace that looks like a person, with a record of everything. If you want the background on why pacing and limits matter, I wrote more in cloud-based LinkedIn automation.

What an AI agent can do with a LinkedIn MCP server

The Apex MCP has 47 tools. They follow a simple loop: see (find people and posts), decide (enrich, check history, judge fit) and act (comment, connect, message), plus the automation and scheduling tools LeadShark was built on.

Plan access matters here, so I've marked it per row. Pro+ is the MCP baseline; most tools that read or act on other people are Apex.

Capability	Example tool(s)	Plan
Search people or companies	`search`	Apex
Search posts by keyword, by anyone	`search_posts`	Apex
Find fresh lead-magnet posts in your niche	`discover_lead_magnets`	Pro+
Read a person's recent posts, comments and reactions	`list_person_posts`, `get_person_activity`	Apex
Turn a post's commenters and reactors into a lead list	`list_post_engagers`	Apex
See your hottest inbound leads	`list_signals`	Apex
Enrich a person or company profile	`enrich`, `enrich_company`	Pro+ (email lookup is Apex)
Pre-DM briefing: your shared history with someone	`get_lead_activity`	Apex
Comment on or react to a post as yourself	`comment_on_post`, `react_to_post`	Apex
Reply to or like a comment	`engage_with_comment`	Apex
Send a connection request	`send_connection_request`	Apex
Accept or ignore inbound invitations in bulk	`manage_invitations`	Pro+
Send a DM (or InMail, using your own credits)	`send_message`	Apex
Read your inbox and a conversation	`list_recent_messages`, `get_messages_with_person`	Apex
Schedule, edit, shift or cancel posts	`schedule_post_with_automation`, `edit_scheduled_post`, `shift_scheduled_posts`, `cancel_scheduled_post`	Pro+
Create and manage comment-to-DM automations	`create_automation`, `edit_automation`, `list_automations`	Pro+
Build lead pages and pull captured emails	`manage_lead_pages`, `get_page_leads`	Pro+
Pause DMs instantly	`set_daily_dm_limit`	Pro+
Check remaining limits, pause or tighten any channel	`manage_activity_limits`	Apex
Read the agent's own action log	`list_actions`	Apex

A couple of honest notes. Enrichment runs through your own connected LinkedIn account and is limited to roughly 250 lookups a day, so save it for leads worth pursuing. And send_message sends a normal DM only to 1st-degree connections; reaching anyone else means an InMail, which spends your LinkedIn plan's paid credits.

The full, current tool reference with inputs and outputs is in the MCP docs.

Example prompts for your agent

These are the kinds of requests that work well. You don't need tool names; the assistant figures that out.

Who engaged with my last post? Build me a lead list, warmest first.

Find recent LinkedIn posts about hiring SDRs and like the three most relevant.

What has Jane Doe been commenting on lately? Draft a connection note
that references it, and check we haven't talked before.

Show my pending connection requests. Accept the founders, ignore the rest.

Schedule my guide post for Tuesday 9am and DM anyone who comments "guide".

I'm off next week. Push all my scheduled posts back 7 days.

What have you already done today? Anything refused or waiting for approval?

Set my daily DM limit to zero.

The comment-to-DM flow in the middle is the same lead-magnet automation the dashboard runs. If that's new to you, here's how turning post comments into automated DMs works.

How to connect a LinkedIn MCP server to Claude

Claude.ai connects over OAuth. There's nothing to install and no API key to copy.

In Claude.ai, open Customize → Connectors, click the + next to Connectors and choose Add custom connector.
Name it (for example "LeadShark Apex") and paste the server URL: https://apex.leadshark.io/mcp. Leave the optional OAuth fields blank and click Add.
Click Connect. A LeadShark sign-in opens; log in, review the requested scopes (mcp:read and mcp:write) and click Approve.
Open the connector's Configure view. Set the read-only tools to Always allow and leave the write tools on Ask every time.
Test it: "Show me my last 3 LinkedIn posts."

How to connect it to ChatGPT

ChatGPT also uses OAuth, but custom MCP servers live under Apps, not Connectors.

In ChatGPT, open Settings → Apps and turn on Developer mode (under Advanced). Then click Create app.
Name it, set Connection to Server URL and paste https://apex.leadshark.io/mcp. Set Authentication to OAuth, tick "I understand and want to continue" and click Create.
Click Sign in with LeadShark Apex, review the scopes and click Approve.

How to connect it to Cursor or Claude Code

For Cursor and Claude Code you run the LeadShark MCP package locally with an API key from your dashboard settings. You'll need Node.js 18 or newer.

In Cursor, create .cursor/mcp.json:

{
  "mcpServers": {
    "leadshark": {
      "command": "npx",
      "args": ["-y", "@leadshark/mcp-server"],
      "env": {
        "LEADSHARK_API_KEY": "${LEADSHARK_API_KEY}"
      }
    }
  }
}

Restart Cursor after adding it. In Claude Code, one command does it:

claude mcp add leadshark \
  --transport stdio \
  --env LEADSHARK_API_KEY=your_key \
  -- npx -y @leadshark/mcp-server

Whichever client you use, the account needs Pro+ or Apex. On any other plan the server refuses tool calls.

Which plan do you need?

Pro ($39/month) covers LeadShark's lead-magnet automations in the dashboard, but not the MCP.
Pro+ ($59/month) connects the MCP with the baseline tools: automations, scheduling, lead pages, invitation triage, lead-magnet discovery and enrichment.
Apex ($119/month) adds the full agent loadout: search, activity and engager reads, inbox, signals, and acting as yourself (comments, reactions, connections, DMs), plus the limits dashboard and action log.

New accounts get a 7-day free trial with no card. On top of that, any account not already on Apex can unlock a one-time, 72-hour Apex trial from Settings to try the full MCP with their own assistant. Compare plans on the pricing page.

Try it with your own agent

If you already live in Claude, ChatGPT or Cursor, the quickest way to judge whether a LinkedIn MCP server is worth it is to connect one and ask it something real. Create a LeadShark account, start the 72-hour Apex trial, paste the server URL into your assistant, and ask it who engaged with your last post. Keep the write tools on "ask every time" for the first week and watch what it proposes.

FAQ

Does LinkedIn have an official MCP server?

As of October 2026, LinkedIn doesn't offer an official MCP server for acting from your personal profile. Every LinkedIn MCP server available is built by a third party. When comparing them, look at how they treat your account: limits, pacing, approvals and logging.

Can I use a LinkedIn MCP server with Claude?

Yes. In Claude.ai, add a custom connector under Customize → Connectors, paste https://apex.leadshark.io/mcp, and sign in with LeadShark to approve access. For Claude Code, install the @leadshark/mcp-server package with your API key. The account needs Pro+ or Apex.

Is using an AI agent on LinkedIn safe for my account?

It depends on what controls sit between the agent and your account. The Apex MCP keeps every write inside your own daily and weekly limits, paces actions, respects off-hours, supports an approve-first mode and has an Emergency Stop. No automation is risk-free, but a governed setup behaves far more like a person than an agent with unlimited access.

Can the agent send messages without my approval?

Only if you let it. You can put the account in approve-first mode so every write waits for your review, and in Claude you can require confirmation for each write tool. Even without those, sends count against the same daily DM limit as your automations, and archived leads are refused outright.

What's the difference between Pro+ and Apex for MCP?

Pro+ connects the MCP and covers your own content and pipeline: automations, scheduling, lead pages, invitation triage, discovery and enrichment. Apex adds tools that read and act on other people: search, engagers, activity, inbox, and commenting, connecting and DMing as yourself.

Do I need an API key?

Not for Claude.ai or ChatGPT; both connect through OAuth with a sign-in and approval screen. Cursor and Claude Code use the local @leadshark/mcp-server package, which needs an API key from your LeadShark settings.