LinkedIn MCP server: post, read and reply from Claude or ChatGPT
A LinkedIn MCP server turns your LinkedIn account into tools an AI agent can call: write a post, publish it, comment as you. What it cannot do is read the comments on that post or touch your inbox, because LinkedIn reserves those permissions. Everything below was checked on October 15, 2026.
MCP is the Model Context Protocol: one standard way for a model to discover and call tools. Point Claude, ChatGPT or Cursor at an MCP server and the tools appear in the session. The protocol is not the hard part here. LinkedIn's permission model is.
What a linkedin mcp server can do through the official API
Any MCP server reaching LinkedIn compliantly is calling the same endpoints you could call yourself. The ceiling is therefore the same: whatever w_member_social grants, and nothing more.
| Action | Endpoint | Permission | Self-serve |
|---|---|---|---|
| Publish a post | POST /rest/posts |
w_member_social |
yes |
| Comment on a post | POST /rest/socialActions/{urn}/comments |
w_member_social_feed |
yes |
| Like a post | POST /rest/socialActions/{urn}/likes |
w_member_social_feed |
yes |
| Read your own posts | GET /rest/posts?q=author |
r_member_social |
no, restricted |
| Read comments on your post | GET /rest/socialActions/{urn}/comments |
r_member_social_feed |
no, restricted |
| Send or read a DM | POST /v2/messages |
Messages API | no, partners only |
| Post as a company page | POST /rest/posts |
w_organization_social |
no, review required |
The Getting Access page states it plainly: "Open Permissions are the only permissions that are available to all developers without special approval," and that set is profile, email and w_member_social. The Network Update Social Actions doc marks r_member_social_feed as "Restricted: granted to select developers only".
So an honest answer to "can an MCP server read my LinkedIn comments" is: not unless its vendor is an approved LinkedIn developer, and most are not. Any server promising it is either approved, or driving your account outside the API, which LinkedIn's User Agreement forbids.
mcp linkedin post: what publishing actually supports
Publishing is the part that works well. Organic post types accepted by the Posts API, from the Posts API reference read today: text only, images, video, documents, articles, multi-image posts, polls and celebrations. Organic carousels are not supported, only sponsored ones.
Two details that bite agents in particular:
- No URL scraping. The API does not fetch a link preview for you. If the agent posts an article, it has to supply the thumbnail, title and description itself, after uploading the thumbnail through the Images API.
- Mentions are positional.
@[Name](urn:li:organization:123)in the commentary only renders as a link when the text matches the entity name exactly, case included. An agent that paraphrases a company name silently loses the mention.
Rate limits on the self-serve product are 150 requests per member per day and 100,000 per application per day. For one person's posting cadence that ceiling never arrives; for a tool serving many members it is the number to design against.
What PlugKit's LinkedIn tools cover, and what they do not
PlugKit is a hosted MCP server over Streamable HTTP at https://api.plugkit.co/mcp, with 103 tools and 3 workflows across eight platforms. LinkedIn is one of them, and its coverage is narrower than Instagram's.
On LinkedIn, PlugKit publishes: text, 1 to 20 images, or one video, to the member's profile, with scheduling through scheduledFor or immediate publication through publishNow. That is what the publishing guide documents as of today. A post queued that way sits in PlugKit, not in LinkedIn's own queue, so if you go looking for it natively, see scheduled posts on LinkedIn has the menu route.
What it does not cover on LinkedIn: comment reading and replies, which the engagement tools serve for Instagram, Facebook Pages and Threads only; the unified inbox, which covers Instagram, Facebook, WhatsApp and Telegram; and analytics, which covers Instagram, Facebook, TikTok, YouTube and X. If your reason for wanting a LinkedIn MCP is an agent that triages LinkedIn comments, no compliant server does that for a personal profile today, PlugKit included.
I would rather write that sentence than have you discover it after paying. The comparison across vendors is in social media MCP servers compared.
claude linkedin mcp: the setup
One command, from the MCP clients doc:
claude mcp add --transport http --scope user plugkit https://api.plugkit.co/mcp
Then connect your LinkedIn account in the PlugKit dashboard with the normal OAuth button, create an API key, and paste it when the client asks for a bearer token. ChatGPT, Cursor and VS Code take the same URL through their own connector settings; the transport is plain HTTP, so nothing needs a bridge or a local process.
Ask the agent for whoami first. It returns the identity and the connected accounts, which is the quickest proof the token works before you let it write anything. The same handshake, with real output, is in the Hermes Agent setup guide.
A read-only week is worth it. Use an OAuth scope of mcp:read if your client supports it, or mark the server untrusted so every write waits for your approval, and test the first posts on an account you do not mind embarrassing.
Should you use MCP or the REST API for LinkedIn?
If the agent writes the copy, MCP. The agent discovers the tools, sees the account list, and publishes in the same turn without you shipping glue code.
If a scheduler or a CMS publishes on a fixed plan, call the REST API directly. No model in the loop means no tokens spent and no paraphrasing risk on mentions. The trade-off in full is in MCP or REST API for your agent, and what Claude can and cannot do on its own is in Claude as your social media manager.
PlugKit costs $49 a month for 5 connected accounts, or $97 for 30, cancel anytime. Seven days free, no card, one connected account during the trial.
FAQ
Is there a LinkedIn MCP server?
Yes, several, and they divide the same way the rest of the LinkedIn tool market does. Servers built on the official API publish and comment as the authorized member. Servers that log into LinkedIn as you can do more, including DMs, and break LinkedIn's User Agreement doing it. PlugKit is in the first group.
Can an MCP server read my LinkedIn DMs?
No. LinkedIn's Messages API is restricted to approved partners, and even they must tie each message to a specific member action, which the docs say excludes automated or scheduled events. No MCP server can compliantly read or send your LinkedIn DMs.
Can Claude post to LinkedIn?
Not by itself. Claude has no network access to LinkedIn and no token for your account. Connect an MCP server that holds the OAuth token, and the publishing tools appear in the session, at which point Claude can draft and publish in one turn.
Does a LinkedIn MCP server work with ChatGPT and Cursor?
Yes, if it speaks Streamable HTTP rather than stdio only. A hosted server is a URL plus a bearer token, which ChatGPT, Cursor, VS Code and Claude Code all accept in their connector settings. A stdio-only server needs to run on your machine, which rules out ChatGPT.
What about posting to a LinkedIn company page from an agent?
That needs w_organization_social, which means applying for the Community Management API and holding an admin role on the page. It is not available self-serve, so most MCP servers, PlugKit included, publish to the member profile only.
One plug for every platform your agent touches
$49 a month for 5 connected accounts, cancel anytime. Every feature in every plan.
Get your API key →