Best LinkedIn MCP Servers in 2026: An Honest Comparison

Search for a LinkedIn MCP server and you get a wall: nine GitHub repositories, three directory sites, a Reddit thread, and a YouTube tutorial. What you don't get is an answer to the question you actually have — which one should I install, and will it get my LinkedIn account restricted?

This is the map. Every tool below was checked on the day of writing: repository status, star counts, authentication model, and whether the thing talks to LinkedIn's API or drives your logged-in session with a cookie. One disclosure up front, because it changes how you should read the rest: we're building a LinkedIn MCP server too, and it isn't finished. Where a tool you can install today is the better answer, this guide says so plainly. (Facts verified 2026-07-28.)

Key takeaways

• LinkedIn publishes no official MCP server, and no official Claude connector. Every LinkedIn MCP server you can install today is third-party.

• They fall into three families: open-source servers from GitHub, aggregator bridges like Zapier and Composio, and hosted servers from LinkedIn content tools.

• The variable that decides your account risk isn't the family — it's whether the server acts through LinkedIn's API with your authorization, or drives your logged-in session with a scraped cookie.

• The most popular option by a wide margin, stickerdaniel/linkedin-mcp-server (star count varies by source, roughly 2.5–3k — check the live repo rather than trust any fixed number), is browser automation, and its own README says automated access may violate LinkedIn's terms.

• None of them fix the writing. MCP is transport. Connecting Claude to LinkedIn does not teach Claude how you write.

Does LinkedIn have an official MCP server? No

There is no LinkedIn-published MCP server, no LinkedIn-published connector for Claude or ChatGPT, and nothing in LinkedIn's developer documentation that exposes its APIs as an MCP surface. The largest community projects say so themselves — stickerdaniel's README carries an explicit disclaimer that the project is "not affiliated with, authorized by, endorsed by, or sponsored by LinkedIn Corporation or Microsoft."

So "official" can't mean "made by LinkedIn." It has to mean something else, and the distinction it actually maps to is the spine of this whole comparison:

• An approved app acting on your behalf. The vendor has a LinkedIn developer application with products like Sign In with LinkedIn and Share on LinkedIn enabled. You grant it permissions through OAuth. It calls documented endpoints with a token you issued and can revoke. Actions are member-authorized, on your own content — the same posture as using that vendor's web app.

• A tool driving your logged-in session. The server takes your li_at session cookie, or your email and password, and operates a headless browser or an unofficial endpoint as if it were you. Nothing is authorized. Nothing is revocable except by changing your password. LinkedIn's User Agreement restricts automated access outside its approved interfaces, and this is squarely that.

Both get described as "a LinkedIn MCP server." They are not the same product category, and the README almost never leads with which one you're holding. Check this first, before stars, before tool count, before anything.

The three families of LinkedIn MCP server

Everything installable today sorts into three groups, and the group tells you most of what you need to know before you read a single line of documentation.

• (a) Open-source servers. Free, transparent, community-maintained. Broad read access — profiles, companies, jobs, inbox, feed. Almost all of them need your session, because there is no public API for most of what they do. A small minority are built on LinkedIn's official API instead, and those look completely different.

• (b) Aggregator and workflow bridges. Composio, Zapier, Apify, Bright Data. LinkedIn is one connector among thousands. Low risk on the publishing side, per-action pricing, and a capability surface shaped by what the platform decided to build rather than by what a LinkedIn creator needs.

• (c) Hosted servers from LinkedIn content tools. The vendor already has approved API access and a product around it, and the MCP server is a second front door to that product. Taplio ships one today. ContentIn's is in build.

LinkedIn MCP servers compared

Every row verified 2026-07-28. Star counts and repository status change fast in this ecosystem — treat anything older than a quarter as unverified.

Family (a): the open-source servers, reviewed

This is the family the search results are made of. Positions one, three, four, six, nine and ten on the query that brought you here are all repositories or directories pointing at repositories. Here's what's actually in them.

stickerdaniel/linkedin-mcp-server

The default answer, and by a distance the most capable. Star counts for this repo swing noticeably depending on which source last crawled it — we saw anywhere from roughly 2.4k to 2.9k stars across different sources checked on the same day, with forks somewhere in the low 400s to low 500s. Treat any specific number you read, including ours, as approximate and check the live repo page. It's Apache 2.0, with commits landing the same week we checked. Per the current README's own tool-status table, it exposes fifteen working tools — get_person_profile, get_my_profile, get_company_profile, get_company_employees, search_companies, search_jobs, search_people, get_inbox, get_conversation, search_conversations, get_feed and more.

Two tools are still listed as in progress, not working: connect_with_person (sending or accepting a connection request) and send_message (sending a reply). Reading your inbox, reading conversations, and searching for people are already working — it's specifically the two write actions that aren't shipped yet. It ships as a Claude Desktop bundle, a Docker image and a uvx command, which is why it's the one every tutorial picks.

Read the mechanism, though. It drives a Chromium browser through Patchright, storing a browser profile on your machine after you log in manually or import an existing session. That is scraping, and the maintainer is straightforward about the consequence: automated access may violate LinkedIn's terms and can lead to account restrictions, personal use only, no warranty, no guarantee of account safety. That honesty is a point in the project's favor. It doesn't change the risk you're accepting.

Note also what it does not do: there is no tool for publishing your own posts, and the two write-adjacent tools it is building — connection requests and message replies — aren't working yet either. It is a research instrument, not a posting instrument. If your reason for wanting a LinkedIn MCP is "post from Claude," this is the wrong tool no matter how many stars it has.

souravdasbiswas/linkedin-mcp-server

Same name, opposite architecture, roughly five stars. It runs against LinkedIn's official API using an OAuth 2.0 flow with the openid, profile, email and w_member_social scopes, and its fourteen tools cover creating and deleting posts, listing your own posts, commenting, reacting, uploading images and creating events. MIT-licensed, with token persistence across restarts.

The catch is setup. You have to register your own LinkedIn developer application, enable Sign In with LinkedIn using OpenID Connect and Share on LinkedIn, configure a redirect URI, and manage the client ID and secret yourself. That's an afternoon of developer-portal work rather than a one-line install. But it's the cleanest illustration in the whole landscape of what an API-backed server looks like: narrower capabilities, dramatically lower risk. If you're technical and you want to publish from Claude without handing anyone a session cookie, this is a genuinely reasonable answer today.

The long tail, and a warning about listicles

Below the top project sit dozens of smaller repositories — eliasbiondo/linkedin-mcp-server (163 stars, people/company/job search with profile scraping), Sharan-Kumar-R/Custom-MCP-Server (95), felipfr/linkedin-mcpserver (77), Linked-API/linkedapi-mcp (63), anysiteio/anysite-mcp-server (62), quinnjr/linkedin-mcp (55), alinaqi/mcp-linkedin-server (53). Capability overlaps heavily. Maintenance does not: some were last touched in early 2025, others yesterday.

Here is the finding that best explains why you should distrust every LinkedIn MCP roundup you read, including the ones published this month. The second-most-cited repository in this space, adhikasp/mcp-linkedin, appears in directory listings, comparison posts and setup tutorials all over the web. On 2026-07-28 it returns a 404, and it is absent from its author's public repository list. Guides are still recommending you install it and hand it your LinkedIn email and password.

Which is the honest summary of family (a): free, transparent, often excellent engineering, genuinely useful for reading data — and a maintenance surface that moves faster than anyone's documentation. If you go open source, verify the repository yourself the day you install it.

Family (b): aggregators and workflow bridges

The aggregators are the least discussed and, for a specific job, the most sensible option in the landscape.

Composio is the strongest of them for LinkedIn specifically: 22 tools, managed OAuth 2.0 with automatic token refresh, and a surface that covers creating posts and comments, deleting posts, reading your profile and network size, company information, share statistics, image upload and registration. It's API-backed, so the risk profile is low, and you never hold a credential yourself.

Zapier's LinkedIn MCP is far narrower — two write actions, create a share update and create a company page update — but it slots into an account you probably already have, exposed at a remote server URL, and every MCP call draws from your plan's task quota. Fine as the last step of a workflow. Not a control surface for your LinkedIn presence.

Apify and Bright Data belong to a different conversation again. Both expose LinkedIn through MCP, both are read-only, and both are extraction platforms: profile search, company employee lists, post scraping, job listings, priced per result or per compute unit. Neither publishes anything. And what they read is largely other people's data, which is the highest-scrutiny activity on this entire list — a different legal question from acting on your own content, and one worth understanding before you point an agent at it.

The rule of thumb: aggregators are good when LinkedIn is one node in a bigger workflow, and poor when LinkedIn is the workflow. If your ask is "when a deal closes in the CRM, post an announcement," Composio or Zapier is the right answer and you can stop reading. If your ask is "help me run my LinkedIn presence from Claude," per-action pricing and a generic connector surface will fight you every day.

Family (c): hosted servers from LinkedIn content tools

Taplio's MCP: live, hosted, and it works

Taplio ships a LinkedIn MCP server today, and it's the most complete hosted implementation currently available. It's a remote server at mcp.taplio.com, added by URL rather than installed, with no API key to manage — it authorizes against your Taplio account the first time a tool touches your LinkedIn data. Taplio documents setup for Claude Code, Claude Desktop and Claude's web connectors, plus ChatGPT, Cursor, VS Code, Windsurf and Zed. Fourteen tools cover inspiration search, draft creation, scheduling, an analytics overview and per-post analytics.

If you already pay for Taplio, this is a straightforward upgrade to how you use it, and we'd rather say that than pretend otherwise.

The honest limitation is architectural, not a knock on the execution: it is a remote control for Taplio's tooling. The drafts it creates are Taplio drafts, produced by Taplio's generation, in Taplio's house style. Moving the interface from a browser tab into Claude changes where you type. It doesn't change whose voice comes out. If Taplio's output already sounds like you, that's a non-issue — and if you're weighing the underlying product rather than the transport, we've written that comparison separately in our Taplio alternatives breakdown.

The ContentIn MCP: in build, with a waitlist

The ContentIn LinkedIn MCP server is not live yet. It's in build, with a waitlist. Everything in this section is what we're building, not what you can install today — every other tool on this page is something you can go and try in the next ten minutes, and ours isn't. We'd rather you know that before the CTA than after it.

The planned tool surface is the LinkedIn workflow rather than a LinkedIn API wrapper: write a post in your voice, generate ideas, repurpose an existing post, schedule and reschedule, list your posts, read your own post and profile analytics, and capture substance from a conversation. It authenticates against ContentIn's approved LinkedIn application — API key at MVP — and works with any MCP client, with quotas shared with your ContentIn account rather than metered separately.

Two design decisions are worth stating, because they're the part we'd defend rather than the feature list.

The first is that publishing requires explicit confirmation, with the exact text echoed back before anything goes live. This sounds like polish. It isn't. An assistant with publish rights and an untrusted document in its context — a pasted email, a fetched web page, a shared PDF — is a prompt-injection incident waiting to happen, and the blast radius is your professional reputation, publicly, under your own name. A publish tool that can fire without a human reading the final string is not a convenience feature; it's an unreviewed deploy to production with your face on it. We think any publish-capable LinkedIn MCP should work this way, ours included.

The second is capture_substance. The interesting thing an MCP server can do isn't pushing text out to LinkedIn — it's pulling raw material back in. Your best material shows up in conversations that were never about content: debugging a pricing objection, arguing through a hiring decision, thinking out loud after a bad customer call. That tool lets any conversation deposit those specifics into the substance bank your posts get written from, so the writing improves because the input improved. Nothing else in this comparison has a loop pointing that direction. If you want the setup mechanics, we've covered what we can honestly cover in our pre-launch getting-started guide, and the waitlist is here.

How to choose: a decision tree

Start with what you're actually trying to do, not with the leaderboard.

• Reading LinkedIn data for research — competitor posts, job listings, company information. Open-source or an extraction platform. Accept that you're in scraping territory, use a throwaway approach where you can, and read the terms question below before you scale it up.

• Publishing as the last step of an automation — a CRM event, a new blog post, a release note. Composio or Zapier. API-backed, low risk, done in an afternoon.

• Publishing from Claude, technical, no vendor — the API-backed open-source route with your own LinkedIn developer app. More setup, no subscription, no cookie.

• Running your personal LinkedIn presence from your assistant — a hosted server from a content tool. Then one question decides it: whose voice comes out the other end? If you're happy with a tool's generated style, take the one that ships today. If the thing you actually want is your own writing, that's a different product, and it's the one we're building.

• Lead generation, bulk exports, mass outreach — nothing on this page is built for that, deliberately. It's the fastest route to a restricted account and the category LinkedIn polices hardest.

What none of them solve

Here's the thing the whole category quietly avoids. Every server above is plumbing. Connect the best one available and the model on the other end still writes like a model: the three-word sentence for emphasis, the manufactured one-line closer, the tidy "it's not X, it's Y" construction that no human has ever said out loud. Faster distribution of writing that doesn't sound like you is not a win. It's the same problem with less friction.

This is why the voice column exists in the table above and nowhere else in this landscape. A server can hand your assistant a publish button; it cannot hand it thirty of your old posts, the phrases you actually reach for, the topics you're credible on, or the record of which of your posts landed. Voice matching is a product problem living upstream of the transport — the difference between an assistant that publishes for you and one that writes as you. It's the same reason a ghostwriter spends the first month interviewing you rather than writing.

Pick your MCP server for the workflow. Judge the tool behind it on the writing.

Frequently asked questions

Is there an official LinkedIn MCP server?
No. LinkedIn publishes no MCP server and no official connector for Claude or ChatGPT. Everything available is third-party. The closest thing to "official" is a server built on LinkedIn's documented API by an approved developer application, acting with permissions you granted.

Will using a LinkedIn MCP server get my account restricted?
It depends entirely on how the server connects. Servers acting through LinkedIn's API with your explicit authorization perform member-authorized actions on your own content, the same posture as using that vendor's web app. Servers that automate your logged-in session with a scraped cookie fall under the automated-access restrictions in LinkedIn's User Agreement and carry real risk — which the maintainers themselves state in their READMEs.

Do I need to be a developer to use one?
For hosted servers, no: you add a URL in your client's connector settings and authorize. For open-source servers, yes in practice — cloning a repository, managing credentials, and repairing things when LinkedIn changes its front end. The API-backed open-source route additionally requires registering your own LinkedIn developer application.

Which LinkedIn MCP servers work with ChatGPT rather than Claude?
MCP is client-agnostic, so any server can in principle work anywhere — but in practice ChatGPT supports remote servers added as connectors, not local ones you run on your machine. That rules out most open-source options unless you host them yourself, and favors hosted servers with a public URL. Claude Desktop, Claude Code, Cursor, VS Code, Windsurf and Zed all support local servers as well.

Can an MCP server post to LinkedIn without asking me?
Technically yes, and this is the question to ask any vendor before you connect anything. A well-designed publish tool gates on explicit confirmation and echoes the exact text back first. Without that, an assistant with publish rights and an untrusted document in its context can be talked into posting by the document. Check the confirmation behavior before you grant publish access, not after.

Is there a free LinkedIn MCP server?
Yes — every open-source option is free, including both the scraping-based and API-backed projects covered above. The cost isn't money; it's setup time, maintenance when things break, and, for the session-based ones, the account risk you're accepting. The paid hosted options are charging for the API relationship, the reliability and the product behind the server, not for MCP itself.

The bottom line

There is no official LinkedIn MCP server, and there probably won't be one soon. What exists is a fast-moving field of third-party projects that differ far more in risk than in capability, and the single question that sorts them is whether they act through LinkedIn's API with your authorization or drive your logged-in session with a cookie.

If you want to read LinkedIn data, the open-source servers are genuinely good and genuinely risky, and you should verify the repository the day you install it. If you want to publish from a workflow, Composio or Zapier will take an afternoon. If you want a hosted control surface for your LinkedIn presence and you're happy with a tool's generated style, Taplio's is live today and works.

And if the thing you actually want is an assistant that publishes writing that sounds like you — that's the one nobody ships yet, and it's what we're building. It isn't available today. Join the waitlist and you'll get setup instructions the moment access opens. In the meantime, the writing side is the part that's live: ContentIn's AI LinkedIn post generator trains on the posts you've actually written, so by the time the server ships, the voice is already yours.