How to give an AI agent access to LinkedIn
Short answer: pick one of three routes. The official API, a browser the agent drives, or a service that holds your logged-in session and exposes it as an API. The third route is what most working setups use in 2026. Here is what each one can actually do.
Why an agent cannot just read LinkedIn
LinkedIn serves a login wall to anyone who is not signed in. A model with web search turned on fetches a profile URL and gets that wall. This is why an agent that seems to know everything about the public web knows almost nothing about LinkedIn.
Reading a profile takes a logged-in session. Sending a message takes one too. The agent does not have a session and cannot create one. Something has to hold that session for it. That something is the access layer, and choosing it is the whole decision.
The three routes
| Official API | Browser automation | Session broker | |
|---|---|---|---|
| How you get in | Apply to a partner program | Paste a session cookie or log in headlessly | Sign in once on the broker's hosted page |
| People search | Not available | Yes | Yes |
| Read posts | Your own pages only | Yes | Yes |
| Send messages | Limited to approved programs | Yes | Yes |
| Speed per action | Fast | Slow. The page has to load and render. | Fast. It is an API call. |
| Breaks when | Rarely | LinkedIn changes its markup | The broker's session expires |
| Who holds your password | Nobody | Whoever runs the browser | The broker, on its own page |
Route 1: the official LinkedIn API
LinkedIn does publish an API. Access is granted per partner program, and the programs cover advertising, recruiting systems, and managing pages you already own.
There is no people search for third parties. There is no general profile read. An agent built on the official API can post to a company page you administer and can read your own analytics. It cannot find a person, read what they wrote last week, or message them. For outreach work this route is closed.
Route 2: a browser the agent drives
This is what most open-source LinkedIn MCP servers on GitHub do. The server runs a headless browser, loads your session, navigates to a page, and reads the rendered HTML.
It works, and it is free. Three things to know before you build on it. The page has to fully load before anything can be read, so a single profile lookup takes tens of seconds. The parsing is tied to LinkedIn's markup, so a layout change silently breaks it. You hand your session cookie to the process, so wherever that process runs is now holding your account.
Route 3: an authenticated session broker
A broker holds a live logged-in session for you and exposes it as an API. You sign in once through the broker's own hosted page. After that every read and every action runs as you, through that session, inside your own account's rate limits.
The difference you feel is speed. There is no page to render, so a profile read returns in about a second and a full drafted action in a few seconds. The difference you do not feel is stability. The broker maintains the session and absorbs the churn, so a LinkedIn redesign is their problem.
Which route to pick
- You manage a company page and want to schedule posts. The official API is enough, and it is the most durable option.
- You are experimenting and do not mind maintenance. An open-source browser-driven MCP server gets you reading profiles today for nothing.
- You want an agent doing daily work on your account. Use a session broker. Speed and stability both matter once this runs every day.
Questions people ask next
Is there an official LinkedIn MCP server?
No. LinkedIn does not publish, endorse, or support a Model Context Protocol server, and nothing in its developer documentation exposes its API as an MCP surface. Every LinkedIn MCP server you can install in 2026 is built by a third party and runs on a session you authorize.
Will this get my LinkedIn account restricted?
Reading is low risk. Sending is where accounts get limited. The pattern that causes trouble is high volume of identical messages sent quickly. Keep daily numbers low, approve each outgoing message yourself, and use a tool that records what it sent so you can see your own pace. You are responsible for staying inside LinkedIn's terms.
Do I have to hand over my LinkedIn password?
You should not have to give it to the tool itself. A broker takes your credentials on its own hosted page and returns an opaque handle to the tool. The tool stores the handle. It never sees the password. Ask any vendor where the password is entered and what they keep afterward.
What can the agent do once it has access?
Search for people by description, read their profiles and recent posts, draft and send connection requests, send messages and replies, comment on posts, and publish posts of your own. Whether it does any of that without asking you first is a setting, and it is the most important setting in the tool.
How do I know what it did?
Ask for an activity log before you connect anything. A good one records every attempt, the exact text sent, the time, and the failures with their reasons. Without a log you are trusting the chat transcript, and the transcript is what the model believes rather than what happened.
How Iridium does it
Iridium is a LinkedIn MCP server built on route three. LinkedIn connectivity is brokered by Unipile. You connect your own account through Unipile's hosted flow, so your password is entered on their page and Iridium only ever holds an opaque handle.
Your agent gets tools for searching people, reading profiles and posts, drafting messages, commenting, and publishing. Every one of them produces a draft that waits for your approval. Approving is a separate step and it is the only thing that sends. Every attempt is written to an append-only log the agent can read back to you.
A typical tool call returns in about six seconds, because there is no browser rendering a page. Connect it in Claude at https://api.iridiumhqmcp.com/mcp, or use the web app if you would rather have a dashboard.
Give your agent LinkedIn access in five minutes
7-day free trial. No credit card. Cancel anytime.
Start free → Read the FAQ