How to Connect LinkedIn to an AI Assistant

“Connect LinkedIn to AI” sounds like one thing. It is really two, and they have almost nothing in common except the word LinkedIn. One of them is sanctioned and fairly boring. The other is the reason people’s accounts get restricted. Before you pick a tool, it is worth knowing which of the two you are actually buying.

What does “connect LinkedIn to AI” actually mean?

When people search this, they usually want one of these:

• Reading — “let the AI look at profiles, my feed, comments, other people’s posts, or search results”.
• Publishing — “let me write a post by chatting, then have it appear on my LinkedIn”.

Those two needs are met by completely different machinery. Reading LinkedIn at scale is not something LinkedIn opens up to ordinary developers, so most tools offering it get the data by pretending to be your browser. Publishing, on the other hand, has an official, free, self-serve permission that any developer can apply for. So the honest answer to “can AI use my LinkedIn” depends entirely on which half you mean.

Why is reading LinkedIn the risky half?

LinkedIn does not hand out a general read permission. The one that would let an app read the posts and activity on your account, r_member_social, is closed — a self-serve developer cannot get it. Analytics, comments and company-page reading sit behind LinkedIn’s partner-only Community Management API, which requires a registered legal company and an application review.

So how do the reading tools do it? Almost always by scraping — automating a logged-in browser, or using a copied session cookie (the small string your browser holds that proves you are already signed in) to fetch pages as if it were you. The tools themselves are usually upfront about the consequence. The most-linked open-source LinkedIn MCP server on GitHub warns in its own README that accounts “can be restricted or banned”, and states plainly that it “does not publish or create posts on your behalf”.

Data vendors such as Apify and Bright Data offer LinkedIn endpoints for extraction rather than posting. That is a different commercial arrangement from a personal tool riding your own session, but it is still firmly the reading side of the split. If reading is genuinely what you need, go in knowing which trade you are making.

Why is publishing the sanctioned half?

Publishing has an official route, and it costs nothing. Two LinkedIn products are open to any developer with no partner status and no payment:

• Sign In with LinkedIn (OIDC) — proves who you are, so the tool knows which account it is acting for.
• Share on LinkedIn — grants the scope (a named permission) w_member_social, which allows publishing a post on your behalf and nothing else.

You grant these on LinkedIn’s own page. You never type your password into the tool. The tool receives an access token — a limited key tied to that one app, that one member, those named permissions — and the grant is then listed in your LinkedIn settings, where one click removes it. Posting happens through LinkedIn’s documented endpoints: POST /v2/ugcPosts, and the newer versioned /rest/posts for multi-image and document posts.

That is the whole difference. One route borrows your identity wholesale; the other asks LinkedIn for a narrow, revocable permission and stays inside it.

What is the piece in the middle called?

The AI app does not talk to LinkedIn directly. Something sits between them, and the current standard for that something is MCP — the Model Context Protocol, an open standard for letting AI assistants call outside tools. An MCP server publishes a list of actions (“publish a post”, “schedule a post”), the assistant reads that list, and calls one when your instruction matches.

You will meet the same idea under three names, which is a large part of why this topic feels confusing:

• MCP server — the technical name for the thing being connected to.
• Connector — what Claude calls it inside its own settings.
• Plugin — the older, looser word people carry over from other software.

A remote or hosted MCP server is one that already runs on the internet, so you add it by pasting a URL. A local one you install and keep running on your own machine. For a normal LinkedIn workflow, hosted is the practical choice — nothing to install, nothing to babysit.

Which AI assistants can connect to LinkedIn today?

This is where a lot of articles start inventing menu paths. We will not. The table says what we have tested and what we have not.

That last row is deliberately unhelpful in one specific way. Support for outside tools inside AI apps changes month to month, and an article that invents a settings path for an app it never opened is worse than one that says “check the docs”. Here is the useful part: if your app can add a remote MCP server or a custom connector, the LinkedIn side works the same everywhere, because the LinkedIn permission lives with the connector rather than with the AI app. Whether your app can add one is a question only that app’s own documentation can answer today.

For the two we have confirmed, there are step-by-step pages: connecting LinkedIn to Claude and connecting LinkedIn to ChatGPT.

What does the setup actually look like?

For a hosted connector such as Kalovio, the shape is the same in both apps:

• Open your AI app’s connector or custom-tool settings.
• Paste the connector address. For Kalovio that is https://kalovio.com/mcp.
• Sign in. LinkedIn shows its own consent screen listing the permissions being asked for. Approving it creates your account and connects LinkedIn in one step.
• Ask the assistant to write a post. It shows you the exact text before anything is sent.

There is no download, no cookie to hunt for in browser settings, no browser extension, and no password given to anyone except LinkedIn.

What can you actually do once it is connected?

On the publishing side, more than most people expect. Kalovio exposes 26 actions, including: publish now, schedule for later, run a weekly posting queue, keep drafts, post one image, post 2–9 swipeable images, post a PDF or document carousel, tag company pages and other Kalovio users so the mention is clickable, save a writing-voice profile so posts sound like you, track clicks on links inside your posts, and delete a post it published. Posts cap at 3,000 characters, files at 20 MB each, and the caption is optional when a file is attached.

Every action that changes something on LinkedIn is two-step. The first call returns a preview plus a signed token covering the exact content; nothing publishes until a second call arrives with your confirmation and a matching token. That gate exists because AI assistants read web pages and documents, and hidden instructions buried inside them are a real attack — prompt injection. Requiring you to approve the exact wording is the practical defence against it.

What can it still not do?

The honest list, because these are the things people otherwise discover on day three:

• It cannot read posts it did not create. That needs the closed r_member_social permission.
• There is no engagement analytics. Impressions, likes and comment data need the partner-gated Community Management API. Link-click tracking is available, and it is not the same thing.
• You reconnect periodically. Member tokens last around 60 days and self-serve apps get no refresh token, so you re-approve when one lapses.
• Company tagging needs the page’s numeric id. Kalovio keeps a shared directory, so a company tagged once resolves by name afterwards.
• Multi-image and document posts use LinkedIn’s newer versioned interface, which may not be switched on for every self-serve app. A refusal is reported clearly, so try it on your own account.
• Document and PDF posts cannot be scheduled yet — immediate publish only.
• No direct messages, no connection requests, no lead scraping. None of that is in scope.
• Chat image attachments do not reach a connector. Neither Claude nor ChatGPT can hand a connector an image you dropped into the chat, so images go through an upload panel or a browser upload link instead.

Is connecting LinkedIn to AI allowed?

Where should I start?

If you mainly want to write and publish, start with whichever app you already use. Claude works on its free plan and is the shortest path from nothing to a published post. ChatGPT works too, but only on a paid plan, only on the web, and you have to switch the connector on in each conversation. If you mainly want to read and analyse other people’s LinkedIn activity, understand first that you are shopping in the scraping half of this market, and read the tool’s own warnings before connecting it to an account you care about.

Kalovio is free while it is in public beta, with no card required. Pricing may change later, and we would rather say that now than surprise you.