LinkedIn MCP Server for Posting From Claude and ChatGPT

A LinkedIn MCP server is a small program that gives an AI assistant tools for LinkedIn. Claude, ChatGPT, Claude Code and Cursor speak the Model Context Protocol, so once a server is connected the assistant can call tools such as "create a post" or "upload an image". You ask in plain words, the assistant picks the tool, and the server talks to LinkedIn.

Most people who search for a LinkedIn MCP server want one of two things. The first group wants to post: turn a release note or a rough idea into a LinkedIn post, with a picture, without opening LinkedIn. The second wants to read: look up profiles, search jobs or pull messages. The two need very different servers, and the difference decides whether you use LinkedIn's own API or something that pretends to be you in a browser.

This page is about the first group: a LinkedIn MCP server for posting. It covers what LinkedIn lets a tool publish, what the popular GitHub projects risk, and how to set up a Claude LinkedIn post or a ChatGPT LinkedIn post step by step, with Spreadcast as the worked example.

If you want the wider field first, our roundup of the best social media MCP servers compares every scheduling vendor that runs one, and the Spreadcast social media MCP server page covers the server across all the networks it supports. This page stays on LinkedIn.

No, not on October 16, 2026. LinkedIn publishes no MCP server, no MCP address to paste into Claude and no LinkedIn connector for posting in Claude's connector settings. A search of LinkedIn's developer documentation on Microsoft Learn for MCP and Model Context Protocol returns nothing about a server. Every LinkedIn MCP server you can install or connect today is built by someone other than LinkedIn.

LinkedIn does have a presence inside ChatGPT, and it is easy to mistake for one. LinkedIn's app in the ChatGPT plugin directory lists LinkedIn as its developer and describes itself as a way to look up professionals: a person's title, company, location and follower count, with a link through to their profile. It does not write or publish posts. So if you came here hoping for an official "post to LinkedIn" button inside ChatGPT, it does not exist yet.

What LinkedIn does offer officially is its API. The LinkedIn Posts API reference on Microsoft Learn documents how an approved tool creates text, image, video, document, article, poll and multi image posts on behalf of a member or a company page. Any MCP server that posts to LinkedIn within LinkedIn's rules is, underneath, a client of this API.

That gives you a first test for any LinkedIn MCP server: ask how it reaches LinkedIn. If the answer is "through LinkedIn's API, after you sign in on LinkedIn and approve a permission", it is working inside LinkedIn's rules. If the answer is "it uses your browser session" or "paste your li_at cookie here", it is not, and the section on account risk below explains what that means.

Yes to both, with limits. LinkedIn has a public API, but only a small part of it is open. LinkedIn's getting access page says most permissions and partner programs need explicit approval, and open permissions are the only ones any developer gets without it. There are three: profile and email from Sign in with LinkedIn, and w_member_social from Share on LinkedIn, which posts, comments and likes for the member who signed in. So the open API for posting covers your own profile and nothing more.

Everything else sits behind an application. The Advertising API, the Community Management API for company pages, the Sales Navigator permissions and the hiring integrations each have their own review, per the same page. There is no open API for searching jobs or reading other people's profiles, which is why the GitHub servers that do those things drive a browser instead.

On price, LinkedIn's API Terms of Use, last revised December 13, 2022, say LinkedIn does not charge a fee for the APIs today but may choose to in the future. The same terms set a default ceiling of 100,000 calls a day, and among the things an app must not do they list using the APIs to automate posting on LinkedIn. That is one more reason a LinkedIn MCP server should put every post in front of a person before it goes out.

A free API does not make every LinkedIn MCP server free. Open source servers cost nothing but your time and a developer app of your own. Hosted servers charge for the service around the API: Spreadcast saves drafts on the Free plan and schedules and publishes on Pro, as the plans table at the end of this page shows.

Not from LinkedIn. A search of LinkedIn's developer documentation on October 16, 2026 found no MCP server for Campaign Manager, the ads side of LinkedIn. Servers that read or change LinkedIn ad campaigns from Claude are built by third parties on the Advertising API, which LinkedIn's getting access page says developers must be approved for before they can use it.

Spreadcast does not run ads. It makes organic posts. TikTok, by contrast, runs an official ads MCP server, covered on our TikTok MCP server page.

The biggest difference between LinkedIn MCP servers is whose name the post goes out under. Posting as yourself and posting as a company page are separate permissions with very different paths to get them.

Posting as yourself is the easy one. LinkedIn's getting access page lists w_member_social under its open permissions, the only ones "available to all developers without special approval". That is why nearly every open source LinkedIn MCP server you find on GitHub posts to a personal profile and nothing else.

Posting as a company page is harder. The w_organization_social permission belongs to the Community Management API, and LinkedIn's page on access tiers says every app starts in a Development tier with a limit of 500 API calls per app and 100 per member in 24 hours, meant for building and testing within twelve months. Production use needs a separate Standard tier application, and LinkedIn says it may decline a partner even when the minimum requirements are met. On top of the app's permission, the person signing in must hold a page role such as Administrator or Content Admin, per the Posts API reference.

For a small app studio the founder's profile is often the better place anyway. The LinkedIn post generator page covers what to write for an app release on LinkedIn.

The swipeable slide posts people call LinkedIn carousels are, technically, documents. The LinkedIn Posts API reference has a content type named Carousels, but marks it as sponsored only and says plainly that an organic carousel is not supported. The organic version is a document post: a PDF, PowerPoint or Word file that LinkedIn shows page by page. The Documents API reference caps those files at 100 MB and 300 pages.

The API has a third option that is easy to confuse with both: the multi image post. The MultiImage API reference describes it as a post with 2 to 20 pictures, uploaded one by one as images. LinkedIn treats it as a separate content type from a document, so a set of slides posted as pictures is not the same post as the same slides exported to one PDF.

So when an assistant offers to post a carousel, ask which of the two it will make. A server that uploads documents can post a real document carousel if something produces the PDF. A server that only uploads images posts your slides as a multi image post.

The exact sizes for each, how to export a sharp PDF from Canva, PowerPoint, Keynote or Figma, and the safe zones that stop text from being cropped are on our LinkedIn carousel size guide. Slide by slide structures you can copy, including ones built around an app, are on our LinkedIn carousel examples page.

LinkedIn is unusually clear about automation on its website. LinkedIn's help page on prohibited software and extensions says it does not permit third party software, including crawlers, bots and browser extensions, that scrape, modify the appearance of or automate activity on its website. It points to section 8.2 of its User Agreement and says members who use such tools risk having their accounts restricted or shut down.

A browser session MCP server is exactly that kind of tool, whatever its README calls it: it drives the LinkedIn website while signed in as you. If the account is restricted, the cost lands on you, not on the project. For a founder whose profile is their main sales channel, that is a high price for saving a few minutes a week.

There is a security side too. A cookie such as li_at is a full key to your LinkedIn session. Anything that can read the config file it sits in gets your access, messages included. The MCP security best practices warn that local servers run with the same privileges as the client, can have direct access to your system and can be reached by other processes on your machine. That is a poor place for a credential that cannot be limited to "post only".

Posting through the official API is not a promise of reach either. LinkedIn's help page on posting updates says it may limit the visibility of posts when it detects excessive post creation that points to inauthentic activity, or the use of an automation tool. So post at a human pace and read each post before it goes out. A server that saves drafts for you to approve fits that better than one that publishes the moment the assistant decides to.

Searches for a LinkedIn MCP for Claude Code, Claude Desktop or Cursor, or for a Docker image, come from people who want the server next to their code. Open source LinkedIn servers fit there, because they run as a program on your machine or in a container you start. Claude Code's MCP documentation says a server you add stays private to one project by default, can be shared with a team through a .mcp.json file at the project's root, or can be added for all your projects. Keep a LinkedIn server out of the shared file if its settings hold your client secret or tokens.

A Docker image changes where the server runs, not how it reaches LinkedIn. A cookie based server in a container still drives LinkedIn as you, and a container that holds your tokens is still a key to your account.

Spreadcast needs no program on your machine in either tool. In Claude Code you add spreadcast.app/api/mcp as a remote server and sign in through /mcp, or use a key from Settings, AI Assistants, which saves drafts only unless you switch on Can Publish Posts. The exact commands for both editors are on the social media MCP server page.

Spreadcast is a hosted LinkedIn MCP server for posting, built for people who make apps. You connect LinkedIn once in Spreadcast and Spreadcast once in your assistant. The difference from a plain posting server is that Spreadcast also makes the post: it designs the picture or the carousel in your app's brand, writes the LinkedIn caption in your brand's voice, and saves the result as a draft for you to approve.

On the LinkedIn side, Spreadcast signs in through LinkedIn's official OAuth flow and asks for three permissions: openid and profile to know who you are, and w_member_social to post for you. Posts go out from the personal profile you connected. It does not use cookies, a browser session or your LinkedIn password, and it does not post to company pages today.

A Spreadcast LinkedIn post can be text alone, text with one picture, text with 2 to 9 pictures, or text with one video of up to 10 minutes. Captions are checked against the 3,000 character limit before anything is saved, and the characters that mean something in LinkedIn's little text format, such as parentheses, are escaped so a caption with "(beta)" in it is not cut short.

The pictures come from the same design code as the Spreadcast app. A single post comes in one of 11 kinds, such as feature, launch, stat or tutorial, in your app's colors, fonts and logo, with a real screenshot of your app in an iPhone frame when your brand kit has one. A carousel is 3 to 8 slides at 1080 by 1350 pixels in one look: a cover, screens of the app and a last slide that sends people to it. On LinkedIn, Spreadcast publishes those slides as a multi image post. It does not export a PDF, so if you want the document carousel format, download the slides and build the PDF yourself.

The caption is written for LinkedIn rather than copied from Instagram. Spreadcast's instructions for LinkedIn ask for a first line that makes one specific statement, then the point in two or three short paragraphs or a short list, then one closing question, in the team's first person, under 1,300 characters and with three hashtags. It never invents numbers, quotes or features that are not in your notes or your app's listing.

Then it stops. Every new connection saves drafts. A draft waits on your Spreadcast calendar until you approve it, and only then does Spreadcast publish it at the planned time. If you choose to let the connection publish as well, the assistant is told to show you the post, the network and the time and ask before it schedules anything. Scheduling and publishing need the Pro plan.

LinkedIn is one of the networks Spreadcast posts to. The same draft can also go to Instagram, TikTok, Facebook, Threads, Pinterest, Bluesky, Telegram and Discord, and YouTube and Mastodon connect too, each with a caption written for that network.

People search for "linkedin mcp server url for claude" because Claude's Add custom connector form asks for one address and nothing else. The honest answer is that there is no single LinkedIn MCP server URL. LinkedIn does not publish one, so any address you paste belongs to whoever runs that server, and that company or person will hold a working connection to your LinkedIn account.

For Spreadcast, the address is spreadcast.app/api/mcp, starting with https. You do not need to type it: Settings, AI Assistants in Spreadcast shows it with a copy button. When Claude first reaches it without a sign in, Spreadcast answers that a sign in is needed, and that answer is what makes Claude show the Connect step and open the Spreadcast consent page.

If you run an open source server from GitHub, there is usually no URL at all. Those servers run on your computer and talk to Claude Desktop or Claude Code through a config entry, not an address. Using one in Claude on the web means hosting it on a public https address and protecting it, because anyone who finds an open address can call its tools.

Be careful with addresses copied from forum posts, videos or AI answers. Anthropic's guide tells you to connect only to servers built and hosted by organizations you trust, to review the permissions asked for during sign in, and to watch for servers that change their tools after you connect. The URL decides who sits between your assistant and your LinkedIn account.

Searches for a Claude LinkedIn skill and a Claude LinkedIn connector often come from the same wish, but the two do different jobs. Anthropic's article on skills describes a skill as instructions for how to complete a task, written in Markdown, that Claude loads when the task comes up. Skills work on Free, Pro, Max, Team and Enterprise with code execution switched on, and Team and Enterprise Owners can give one to the whole organization.

For LinkedIn, a skill is the place for your rules: how long your posts run, which words you never use, how you open and close, the three hashtags you always pick. It cannot post. Posting needs a connector, an MCP server with LinkedIn permission. The same article says the two work together: the connector gives Claude the tools and the skill tells it how to use them well.

With Spreadcast you may not need a skill for the basics. The server already offers Claude each network's writing rules and your brand kit as tools, so a skill is best kept for what only you know, such as topics you stay away from or the way you sign off.

It does not post to company pages. Spreadcast asks LinkedIn only for permission to post as the member who connects, so posts go to a personal profile.

It does not post document carousels. Carousels go to LinkedIn as a multi image post, and there is no PDF export. If you want the document format, download the slides and assemble the PDF in the tool you use for slides.

It does not read LinkedIn numbers. Reading a member's posts and their likes needs a permission LinkedIn restricts to approved users, per the Posts API reference, and post and profile analytics belong to the Community Management API, per LinkedIn's access tiers page. So Spreadcast shows no LinkedIn likes, comments or follower counts, and LinkedIn results are best checked in LinkedIn itself.

It does not read profiles, send messages, send connection requests or reply to comments, the jobs browser session servers are built for.