LinkedIn MCP server: what an AI assistant can and cannot do on LinkedIn, Instagram and X

What a LinkedIn MCP server is

A LinkedIn MCP server is a Model Context Protocol server whose tools act on LinkedIn: an AI app such as Claude or ChatGPT calls a tool, and the server makes the call to LinkedIn. The Model Context Protocol is an open standard for connecting AI applications to outside systems, and a server offers those systems to the model as named tools.

An Instagram or X MCP server works the same way for its own network. A social media MCP server puts several networks behind one connection, so one assistant can draft a LinkedIn post and an Instagram caption in the same conversation. What any of them can do is capped twice: by the tools the server offers, and by what each network’s API lets an app do.

What LinkedIn's API lets an MCP server do

LinkedIn’s API lets any app sign a member in and post, comment and like as that member; most other permissions need LinkedIn’s approval. LinkedIn lists profile, email and w_member_social as its open permissions, the only ones available to every developer without special approval (Getting Access to LinkedIn APIs, last updated Jun 26, 2025).

Two rows decide what any LinkedIn MCP server can honestly offer. r_member_social is closed: LinkedIn’s Community Management API FAQ (updated May 15, 2026) says it is not accepting access requests for it at this time, so an app that does not already hold it cannot read a member’s own post history through the API. And the Post API schema (updated Apr 30, 2026) accepts only PUBLISHED as the state of a new post, so an app cannot create a draft or a scheduled post on LinkedIn; a server that schedules has to keep the post and send it at the minute.

What LinkedIn servers in the MCP Registry say they do

Most LinkedIn MCP servers in the official MCP Registry describe reading people, company or job data or working an inbox (37 of 55); 5 describe publishing a member’s own posts. On Oct 28, 2026, a search of the registry for “linkedin” returned 55 distinct servers. We read each server’s own registry description and counted the jobs it names; one server can name several.

None of the 55 sits under a LinkedIn namespace. The registry ties each server name to a GitHub account or a domain its publisher proved they own (MCP Registry authentication), and only that owner can publish under it, so a server from LinkedIn would sit under a LinkedIn namespace. The registry checks who published a server, not what its code does, and leaves security scanning to the wider ecosystem (About the MCP Registry).

The descriptions do not say which servers hold a LinkedIn partner approval. Before connecting one, ask how it reaches LinkedIn: through the API permissions in the table above, or through your own session.

Where LinkedIn draws the line on automation

LinkedIn’s User Agreement prohibits working around the API with copied sessions, scrapers or bots. In section 8.2 of the LinkedIn User Agreement (effective Nov 3, 2025), members agree not to:

• use another person’s account, share log-in credentials or copy cookies;
• use crawlers, scripts, browser plugins or add-ons to scrape profiles and other data;
• use bots to send messages, add contacts, or create, comment on, like or share posts.

The same agreement makes members responsible for what happens through their account, and lets LinkedIn restrict, suspend or terminate an account that breaches it. The account at stake is the member’s own, whichever server did the work.

Instagram and X rules that apply to any MCP server

Instagram and X set their own limits on any app that posts for a person, whichever MCP server sits in front of it. The rules below come from each platform’s developer pages, read on Oct 28, 2026.

overads sets neither AI-disclosure flag today. A post that Instagram or X refuses shows that network’s own error on its delivery record in overads.

What the overads MCP server does on each channel

The overads MCP server drafts, schedules and publishes posts to Instagram, Facebook, LinkedIn, X, Bluesky, Mastodon and Pinterest through each network’s own API. Threads, YouTube and TikTok are built but not live, so it does not post there. The limits below are the ones overads checks when a post is created or edited.

The tools an assistant uses for those posts, from the overads tool list:

Every tool, its parameters and the scope that unlocks it are on the MCP tools reference. For LinkedIn on its own, the LinkedIn post scheduler page covers what overads posts there and the limits LinkedIn documents.

How approval works when an assistant posts

In overads, a post an assistant schedules waits for a person whenever the workspace’s Require approval rule is on, or when the post itself asks for approval.

• Owners and admins switch the rule on in Settings, Rules, Require approval; it is off until someone does. It holds posts made in the app, over the API and MCP, by the overads assistant and by workflows.
• With the rule on, create_post still succeeds, but the post lands as pending with an APPROVAL_REQUIRED warning, and the scheduler does not send it until a member approves it in the app.
• No MCP or API call approves a post; a member does it in overads.
• draft_on_brand_post saves a draft and never schedules or publishes.
• A connection made by signing in shows each permission with its own checkbox, so an assistant can be allowed to read without being allowed to post.

That matches the MCP specification, which says there should always be a human in the loop able to deny tool invocations (MCP specification, Tools). The scheduling and approvals guide has the full rule.

How to connect Claude, ChatGPT or another assistant

Connecting an assistant to the overads MCP server takes one URL and one sign-in:

1. Use a workspace on the Pro or Scale plan. The MCP server is included on both, and a paused workspace cannot connect.
2. In ChatGPT, Claude or Claude Code, add a custom MCP server with the URL https://api.overads.io/public/v1/mcp.
3. Sign in to overads in the window that opens, pick one workspace, untick any permission the assistant should not have, and approve.
4. For Cursor, VS Code, Codex, Windsurf, n8n, Zapier or any other client, create an API key in Settings > API (the API keys and MCP page) with only the scopes it needs, and send it as a bearer token.
5. Ask the assistant which accounts are connected. It calls list_connections and gets the ids it posts to.

Each client has its own setup page under MCP setup for AI assistants, and the MCP server docs cover sessions, errors and disconnecting. The guide to scheduling social media posts with Claude walks through a first post.

Where overads fits

overads fits a team that wants an assistant to draft and schedule its own posts across several networks, with the option to make a person approve every post before it goes out. It is not a LinkedIn data or outreach tool:

• It does not read LinkedIn profiles, search people, send messages or invitations, or run outreach. For prospecting, LinkedIn’s approved route is Sales Navigator and its partner program.
• It does not read a member’s LinkedIn post history, comments, likes or engagement. Over MCP it returns no post engagement on any network today; Facebook and Instagram engagement is built but switched off.
• It does not comment, like or reply on any network.
• It does not set Instagram’s or X’s AI-disclosure flags.
• It does not post to Threads, YouTube and TikTok yet; a scheduler that does fits better there, as the comparison with Buffer sets out.
• Over MCP, ad accounts are read-only: the server reads results from Meta, Google, X and Snapchat and changes nothing in them.
• It needs a paid workspace on Pro or Scale; there is no trial.