LinkedIn MCP: What's Official, and What Logs In as You

LinkedIn, the professional network, does not publish an MCP server you can add to Claude or Cursor. MCP is the open standard that lets an AI assistant like Claude or ChatGPT call another product’s tools directly. Every LinkedIn MCP server you’ll find on GitHub was built by someone else.

The most popular one is stickerdaniel/linkedin-mcp-server, with 3,883 GitHub stars on October 15, 2026. It works by running a browser that’s signed in to your LinkedIn account. Its own FAQ warns that “LinkedIn’s User Agreement prohibits automated access, and accounts using automated tools can be restricted or banned.”

Four of the five most-starred LinkedIn-only MCP servers reach LinkedIn through a browser session on your account. LinkedIn’s official route is far narrower, and for most members it means posting.

Is there an official LinkedIn MCP server?

Not one you can add to Claude or Cursor. On October 15, 2026, we searched LinkedIn’s developer docs on Microsoft Learn for “MCP”, “MCP server” and “Model Context Protocol”. All three came back empty.

What LinkedIn has built is two apps for ChatGPT. Both sit in ChatGPT’s plugin directory with LinkedIn named as the developer.

• LinkedIn: a people lookup. Its listing reads “Look up any professional on LinkedIn without leaving ChatGPT.” You see “their title, company, location and follower count.”
• LinkedIn Ads: a beta that “provides read-only access to ad set performance.” You can ask about ad sets and break results down by date range, ad set or ad.

Neither app is built to read your feed, messages or saved posts, or to publish anything. OpenAI’s own page adds a caveat. “Availability depends on the plugin, your plan, and workspace settings.”

We found no official server for LinkedIn Learning, Sales Navigator or company pages either. Each has an API, and each needs LinkedIn’s approval first.

Claude’s connector directory has nothing from LinkedIn. It does list third-party tools that post to LinkedIn, such as Taplio and Typefully. If MCP itself is new to you, our plain explanation of MCP covers the basics.

What LinkedIn’s own API lets an agent do

This is the API that tools playing by LinkedIn’s rules build on, so it sets the ceiling. LinkedIn’s getting access page is blunt about it. “Most permissions and partner programs require explicit approval from LinkedIn.”

Three permissions are open to every developer, in two products.

• Sign in with LinkedIn: profile and email read your name, headline, photo and primary email address.
• Share on LinkedIn: w_member_social lets an app “Post, comment and like posts on behalf of an authenticated member.”

So an ordinary developer’s app can sign you in and post as you. LinkedIn’s Share on LinkedIn docs set a limit of 150 requests per member per day.

Everything else needs LinkedIn to say yes.

• Company pages: the Community Management API. Approved apps start on a development tier capped at 500 calls a day per app and 100 per member.
• Ads: the Advertising API, by application.
• Sales Navigator: approved Sales Navigator Application Platform partners only.
• Your own posts: reading them back takes r_member_social_feed, which LinkedIn marks “Restricted” and “granted to select developers only.”

None of the open permissions reads your feed, your messages or your saved posts. That’s why the official-API tools below are built around posting. To build a small posting server yourself, see our guide to building your own MCP server.

The EU exception

Members in the European Economic Area and Switzerland get one more door. To meet the EU’s Digital Markets Act, LinkedIn’s Member Data Portability API lets you create an app that fetches your own LinkedIn data.

It reaches your messages, your shared posts and your saved items. But each saved item comes back as “the saved date and URL of a post, article, or other content.” You get the link, not the post, the same gap LinkedIn’s own data export has.

Setup takes a developer app on a default company page LinkedIn provides, plus a token you generate.

How third-party LinkedIn servers get in

Each option below reaches LinkedIn in one of three ways. It drives a browser signed in as you, it calls the official API, or it scrapes pages without logging in. The route decides what it can do, and what you’re risking.

A browser signed in as you

stickerdaniel/linkedin-mcp-server is the big one. It’s free, its latest release shipped on October 10, 2026, and it runs a Chromium browser on your machine. On first use, it “reuses a LinkedIn session from a signed-in local browser if it finds one,” by reading that browser’s LinkedIn cookies.

Read from its source code, its tools include get_feed, get_inbox, send_message, connect_with_person and search_people. It has a tool for saved jobs and none for saved posts.

Here are the next four LinkedIn-only servers by stars, checked October 15, 2026. Three of them work the same way.

• eliasbiondo/linkedin-mcp-server (195 stars): a local browser you log in to once. Its README says “Scraping LinkedIn may violate their Terms of Service.”
• gtm-api/linkedin-mcp (91 stars): a hosted service that runs each account in a cloud browser with its own proxy.
• felipfr/linkedin-mcpserver (90 stars): the exception. It calls LinkedIn’s official API, but it hasn’t been updated since March 28, 2025.
• Linked-API/linkedapi-mcp (70 stars): Linked API’s server. Its FAQ says “You enter your password yourself, directly in your isolated cloud browser.”

Paid services sell the same access from the cloud.

• Unipile: a hosted LinkedIn API with an MCP server included. Its docs connect an account with a username and password, or with “the li_at cookie.” The stickerdaniel README says that project “is supported by Unipile.”
• HeyReach: outreach automation. Its homepage offers to “Automate actions such as connection requests, messages, voice notes, and profile views.” You connect an account by exporting your LinkedIn cookies with the Cookie-Editor Chrome extension and importing them, which its help center calls “the preferred way,” or with your LinkedIn email and password. Its Growth plan includes an MCP server, listed in our MCP directory.
• MCPBundles: an official-API server, plus a Sales Navigator one that acts as your account, up to 500 calls a day.

Some vendors say so plainly. gtm-api’s README says LinkedIn’s User Agreement “does not permit third-party automation, so every tool in this space carries risk and no vendor can honestly promise otherwise.”

LinkedIn’s official API

These tools connect through LinkedIn’s own sign-in screen and stay inside what the API allows, which mostly means posting.

• Upload-Post: its LinkedIn MCP page says “It runs on LinkedIn’s official API” and “You grant access on LinkedIn’s own OAuth screen.” It publishes text, documents, images and video to your profile or a page you manage, and schedules them. The free plan includes 10 uploads a month, and Basic is $24 a month, or $16 billed annually.
• Postiz: an open-source scheduler. Self-hosting means making your own LinkedIn developer app, and its docs warn that without Advertising API approval you “will not have the ability to refresh your tokens.” Its pricing page says “Every plan includes the MCP server, CLI and API.”
• Ayrshare: says it “integrates directly with each social network’s official APIs and partnership programs.” Its MCP server takes an API key, and each LinkedIn account is capped at 150 posts a day and must be re-authorized every year.
• Merge: its Agent Handler’s LinkedIn connector signs in through LinkedIn OAuth, and its listed tools create text, article and image posts.

Scrapers that never touch your account

Apify’s LinkedIn MCP server runs a community actor by Dev Fusion, and its listing promises “No LinkedIn Cookies Required.” It takes profile URLs, tries to find email addresses and costs $10.00 per 1,000 results.

Your account isn’t involved, so the login risk doesn’t apply. But LinkedIn’s agreement bans using software “to scrape or copy” its profiles, and this data includes people’s contact details. Check with your legal team first.

Prices as printed on each vendor’s site, October 15, 2026.

LinkedIn’s rules, word for word

LinkedIn’s User Agreement, effective November 3, 2025, lists what members agree not to do in section 8.2. Three items cover the servers above.

• “use or attempt to use another’s account (such as sharing log-in credentials or copying cookies)”
• “Develop, support or use software, devices, scripts, robots or any other means or processes (such as crawlers, browser plugins and add-ons or any other technology) to scrape or copy the Services, including profiles and other data from the Services”
• “Use bots or other unauthorized automated methods to access the Services, add or download contacts, send or redirect messages, create, comment on, like, share, or re-share posts, or otherwise drive inauthentic engagement”

LinkedIn’s help page on prohibited software and extensions says it doesn’t “permit the use of any third party software, including ‘crawlers’, bots, browser plug-ins, or browser extensions that scrape, modify the appearance of, or automate activity on LinkedIn’s website.”

Then it names the cost. “Any member who uses tools for such purposes is in violation of the User Agreement. This means that they risk having their accounts restricted or shut down.”

Section 3.4 of the agreement backs that with “the right to restrict, suspend, or terminate your account if you breach this Contract.” LinkedIn’s help page on invitation limits gets specific. “If you send an excessive number of invitations and we suspect the use of an automation tool, we may suspend or restrict your account.”

LinkedIn doesn’t say how many invitations count as “many”, so no vendor can promise you a safe volume.

The don’ts also carry a carve-out. They apply “unless otherwise explicitly permitted by LinkedIn in a separate writing,” and apps on the official API run under LinkedIn’s API terms. A server that borrows your session has only your login.

Which route fits what you want

Here’s our take. If a LinkedIn MCP server can read your home feed, it’s using your login, and LinkedIn’s agreement bans that in plain words. Don’t connect one to the account your career runs on, however many stars it has.

The useful version of LinkedIn plus AI is smaller than the search results suggest. Let Claude draft, post through the official API, and paste in anything you want it to read.

Connect LinkedIn to Claude the official way

We’ll use Upload-Post because its free plan covers LinkedIn and its hosted server signs in with OAuth. Postiz works much the same if you self-host.

1. Connect LinkedIn to Upload-Post: create a free account and connect your LinkedIn profile or page. You approve access on LinkedIn’s own screen, so Upload-Post never sees your password.
2. Add the connector: in Claude Desktop, open Customize, then Connectors, then +, then Add custom connector. Name it Upload-Post and paste https://mcp.upload-post.com/mcp as the URL.
3. Sign in: click Connect and sign in with your Upload-Post account.
4. Start with a draft: ask for a draft before anything posts, read it, then tell Claude to publish or schedule it.

In ChatGPT, Upload-Post is a reviewed app in the directory, so you search for it under Apps and click Connect, where your plan and workspace allow it.

Draft in your voice

Below are my last five LinkedIn posts, pasted by me.
Study how each one opens, how long it runs and its tone.
Draft one new post in the same voice that takes the idea
in my most recent post one step further.
Show me the draft. Do not publish or schedule anything.

You paste the posts because no open API permission reads them back. When the draft is right, ask Claude to schedule it through Upload-Post.

Research a company

Below is the text of a company's LinkedIn page and two
of its recent posts, pasted by me. Using only that text,
tell me what they sell, who they seem to be hiring and
what they keep talking about. Then give me three
questions to open a first call. Do not search the web.

Pasting a public page needs no connector, no login and no scraping.

Already connected one that logs in as you?

1. Stop the server: remove it from your AI client. For the stickerdaniel server, the --logout flag clears the stored session.
2. End its sessions: go to Me, then Settings & Privacy, then Sign in & security, then Where you’re signed in, and click Sign out of all these sessions. That skips the session you’re using, so then sign out of LinkedIn in that browser and sign back in.
3. Change your password: if you typed it into a cloud tool, change it now.
4. Remove connected apps: under Settings & Privacy, go to Data privacy, click Change next to Permitted Services, then click Remove next to anything you don’t use.

LinkedIn warns that an app’s owner “may retain your LinkedIn data received prior to the settings change.” Ask the vendor to delete what it holds.

Where ContextBolt fits

ContextBolt is ours, so weigh the bias. Everything above points one way. Let Claude draft, and hand it material by pasting it in. For writing, the best material is what you have already read.

ContextBolt is a Chrome extension that saves any web page with its text in one click, into one library where every save is tagged and grouped into topics automatically. Search in the extension is instant, ranked and runs on your device.

Pro adds an MCP endpoint, signed in with OAuth, so Claude can search the whole library by meaning in the same chat where it drafts your next post. It only knows what you chose to save. When it opens a saved page, it reads the first 2,000 characters.

Try this on Pro

Search my ContextBolt bookmarks for articles I saved
about hiring. Group them into three themes. For each
theme, quote one line from a saved article and give
its link. Then suggest one LinkedIn post I could write
that adds something none of them say. Do not draft it.

Basic is free up to 150 bookmarks. Pro is $6 a month and removes the cap, adds cloud sync and the MCP endpoint.