LinkedIn MCP Servers in 2026: 12 Options Compared by What They Can Actually Do

There is no official LinkedIn MCP server. LinkedIn has not published one, and neither Anthropic nor OpenAI ships a LinkedIn connector. Every "LinkedIn MCP" you can install is third-party, and they do very different things: some publish posts through LinkedIn's API, some read profiles and send messages by driving your browser, and some only extract data.

This page sorts the twelve that come up most, using two sources only: each vendor's own documentation, and LinkedIn's own API documentation on Microsoft Learn. Everything was checked on 26 September 2026, and every row links to where the fact came from.

What LinkedIn's API lets any MCP server do

Before comparing tools, it helps to know what LinkedIn's official API allows. That list decides which MCP servers can be built "properly" and which can only work by automating a browser. This is the part other comparisons skip, and it answers most of the "why can't my AI do X on LinkedIn" questions.

Three practical consequences:

1. Posting is the one job an ordinary app can do legitimately. That is why the API-based servers below are all publishing tools.
2. Company pages need an approved app. A server that makes you register your own developer app will usually only post to your personal profile, because a new app does not have Community Management access.
3. If a server sends DMs, sends connection requests or searches people, it is almost certainly driving a browser. That is not automatically wrong, but it is the category LinkedIn's User Agreement restricts, and it is where the account risk sits.

Pick by job, not by star count

• Publish your own posts from Claude or ChatGPT. Use a hosted server that connects through LinkedIn's API with its own approved app: you sign in once with OAuth and never handle a token. Antwork, Kalovio, 2pr.io, Taplio and PostMCP are in this group.
• Post as the last step of an automation (a closed deal, a new blog post). Composio or Zapier, where LinkedIn is one connector among many.
• Research profiles, search people, read your inbox. stickerdaniel/linkedin-mcp-server (free, local) or Linked API (hosted, paid). Both automate a browser. Read their warnings before pointing them at an account you care about.
• Bulk profile data. Apify runs scraper actors and bills per result. It publishes nothing.

The 12 servers compared

"Not stated" means the vendor's own page does not say. It does not mean the feature is missing. Vendors change fast in this space, so open the linked page before relying on any single cell.

Publishing servers, one by one

Antwork (ours). A hosted MCP server at https://api.antwork.io/mcp. You add the URL to your AI app and sign in with LinkedIn once. It connects through Antwork's own LinkedIn app, which is approved for company pages, so you never register a developer app or paste a token, and you can revoke access from LinkedIn at any time. It posts to personal profiles and company pages, handles images, video and PDF document carousels, schedules posts, adds a first comment under a published post, and reads back impressions, reactions, comments and shares. Before drafting, your AI reads your recent posts on that account so drafts start from how you already write, and every post lands as a draft you can review before it is scheduled or published. It also covers X, Instagram, Threads, Facebook, TikTok, YouTube and Pinterest from the same connection. Limits: it does not send DMs, send connection requests or read other people's profiles, because LinkedIn's API does not allow that for ordinary apps.

Kalovio. A hosted server focused only on LinkedIn, using its own app with Share on LinkedIn. It supports image carousels and document posts (with a note that documents "may be product-gated"), and a posting queue. It is explicit about what it leaves out: "No DMs, no connection requests, no lead scraping", and no real engagement analytics beyond link-click tracking. Free while in public beta.

2pr.io. The broadest tool surface of the hosted publishers: 45 tools, including scheduling, replies, reactions, post stats, best posting times and a searchable library of viral posts. The docs say it is "not a scraping API". It needs a Pro or Team plan.

Taplio. A hosted server at mcp.taplio.com with 27 tools covering drafts, scheduling, comments, analytics and inspiration search, plus media on drafts. A good fit if you already use Taplio, since the server drives your Taplio account.

ContentIn. A hosted server that authenticates with an MCP key you create in ContentIn. It publishes posts with up to five images; its own docs say "GIFs, video and PDFs aren't supported" and it lists comments you wrote rather than posting new ones. Included in the Pro plan.

PostMCP. Runs locally through npx with a PostMCP API key, and connects to LinkedIn through PostMCP's own app, so no developer app is needed. It posts to profiles and company pages with up to 20 images and has a free tier. Using it from claude.ai requires hosting the server yourself.

SocialClaw. A multi-network scheduler with a LinkedIn MCP page. Setup uses a workspace API key, which on claude.ai goes in the server URL. Its page does not say how it connects to LinkedIn.

Composio and Zapier. Composio exposes 22 LinkedIn tools through a managed OAuth app, including company-page posting and share statistics. Zapier's LinkedIn MCP has two actions, create a share update and create a company update. Both are best when LinkedIn is one step in a bigger workflow.

abhineet34/linkedin-mcp-server. The honest do-it-yourself route on the official API: free and open source, but you register your own LinkedIn developer app and paste an access token that expires every 60 days. Scheduling and comments are on its roadmap.

Research and outreach servers

stickerdaniel/linkedin-mcp-server is the most popular LinkedIn MCP on GitHub and the most capable reader: profiles, people search, your feed, your inbox, sending messages and connection requests. It works by driving a browser logged in as you, and its README is direct about the risk: "LinkedIn's User Agreement prohibits automated access, and accounts using automated tools can be restricted or banned." It does not publish posts.

Linked API offers similar reach as a hosted service, running "a dedicated cloud browser on an account you own", including Sales Navigator tools. It comments and reacts but has no create-post tool.

Apify is an extraction platform: profile data and email discovery, billed per result, with no LinkedIn cookies required and no publishing.

If your goal is posting, none of these three is the right tool, however many stars they have.

How to connect a LinkedIn MCP server

The steps are the same for any hosted server. Here they are with Antwork's URL; swap in your vendor's.

• Claude (claude.ai and Claude Desktop): Settings → Connectors → Add custom connector, paste https://api.antwork.io/mcp, then sign in with LinkedIn when asked. The full walkthrough, with what Claude shows you before anything posts, is in how to connect Claude to LinkedIn.
• ChatGPT: turn on Developer mode in Settings → Security and login, open Plugins in the sidebar, click +, and paste the URL. ChatGPT only accepts remote servers, so local GitHub servers will not work here.
• Claude Code: run claude mcp add antwork --transport http https://api.antwork.io/mcp, then /mcp to sign in.
• Cursor: add the URL as a remote MCP server in Cursor's MCP settings.

Then ask something concrete: "Draft a LinkedIn post about what we shipped this week, in the style of my last posts, and schedule it for Tuesday at 9."

Before you give an AI publish rights

• Prefer OAuth over pasted tokens. An OAuth grant can be revoked from LinkedIn's settings. A token pasted into a config file lives wherever that file goes.
• Keep a review step. An assistant with publish rights and an untrusted document in its context (an email, a web page) can be talked into posting. Choose a server that creates drafts and publishes only on an explicit request.
• Know the ceiling. Share on LinkedIn allows 150 requests per member per day. No tool gets around that, and none should try.
• Match the tool to the job. Use a browser-automation server only for work that genuinely needs it, and never on an account you cannot afford to lose.