LinkedIn MCP servers compared
Eight servers on one table. Hosting model, authentication, write actions, pacing controls, and license. Each row comes from that server's own public README or website.
The comparison
Names are the ones each project uses in public. Cells read "Not documented" when the linked page says nothing about that trait.
Checked 2026-10-28.
| Server | Runs | Auth | Write actions | Pacing and safety | Open source |
|---|---|---|---|---|---|
| Crispy | Hosted, remote HTTP | OAuth or a Bearer API key | Search, messaging, content, campaigns, contacts, lists | Daily caps per action category, optional warm-up | Hosted service, server code not published |
| MCP Server for LinkedIn | Self-hosted, drives your own browser session | Your own LinkedIn session in a local profile | Messages, connection requests, conversations | Tool calls run one at a time. No pacing limits documented | Apache 2.0 |
| LinkedIn MCP Server | Self-hosted Node server, calls the official LinkedIn REST API | LinkedIn OAuth access token in your config | Posts, polls, images, comments, reactions | Not documented in their README | MIT |
| mcp-linkedin | Self-hosted, stdio | An API key from a third-party connection service | Publish posts, comment, react | Dry run by default, publish after confirmation | MIT |
| LinkedIn MCP (jmougeot) | Self-hosted, Chrome extension on your own session | Your own LinkedIn session in Chrome | Messages, invitations, profile reads, people search | Daily and hourly quotas, random delays, work hours | Public repository, no license stated in the README |
| Taplio LinkedIn MCP Server | Hosted remote server | OAuth 2.0 in the browser | Posts, comments, likes, connection requests, analytics reads | Queued actions are paced, and their docs say to confirm irreversible actions first | MIT repository, hosted server |
| La Growth Machine MCP | Hosted, remote HTTP | OAuth sign-in with your La Growth Machine account | Campaigns, leads, audiences, inbox replies (50 tools, 26 act) | Writes ask for approval by default, credit cost shown first | Apache 2.0 |
| Kondo MCP | Hosted, Streamable HTTP | Sign in with your Kondo account | Labels, archive, reminders, and reply drafts you review and send yourself | Requests to LinkedIn run one at a time, a few seconds apart | Not documented in their docs |
Also published
- Zapier publishes a LinkedIn MCP Server.
- Dripify publishes an MCP Server for AI assistants.
- The official MCP Registry lists more LinkedIn servers from independent teams. A search on 2026-10-28 found none under a LinkedIn namespace.
How to choose
Three questions separate these servers more than any feature list does.
Whose account does it act on
A hosted server usually acts on an account you hold with that company. A self-hosted server usually acts on your own LinkedIn account. The two behave differently when you leave.
Who runs the server
Hosted means someone else runs the process and the updates. Self-hosted means you run it, and you patch it.
What the agent may write
Read tools are easy to reason about. Write tools are not, so check how each server confirms writes and how it paces them.
Using Crispy as your server
Crispy is a hosted LinkedIn MCP server that acts on your own LinkedIn account. One remote endpoint covers search, messaging, content, campaigns, contacts, lists, and account settings. Pacing runs through daily caps per action category, with optional warm-up.
Add https://crispy.sh/api/mcp to your client, then follow its setup guide. Authentication is OAuth or a Bearer API key.
These controls do not guarantee against restriction. LinkedIn's User Agreement prohibits unauthorized automation, and the caps do not authorize the activity.
Questions about LinkedIn MCP servers
Is there an official LinkedIn MCP server?
Every LinkedIn MCP server in this comparison is built by another team. The MCP Registry lists none under a LinkedIn namespace. LinkedIn's developer portal links a Develop with MCP entry, so read LinkedIn's own page for its current status.
Which LinkedIn MCP server should I use?
It depends on whose account the server acts on, who runs it, and what the agent is allowed to write. The table above compares those traits for each server.
Are these servers hosted or self-hosted?
Both. Some run as hosted services, some run on your own machine. The Runs column shows which is which for each row.
Do these servers support write actions?
Most do. The table lists what each server documents for writes, and what it documents for pacing and confirmation.
Keep reading
All your LinkedIn growth in one place.
Create your account and connect LinkedIn. Then ask Claude or ChatGPT.