LinkedIn MCP servers

This is the reference implementation of the category, and by star count the one most people end up using. It does not call an API: it drives a real Chromium through Patchright on your own machine, logged in as you, and parses what LinkedIn renders. That design decision explains both its reach and its risk. Because it sees what a signed-in human sees, it reaches places an API never will — the home feed, global post search, your message threads, the employee list of a company. Because it is a browser doing it, LinkedIn's page structure is the contract, and the project's answer is to ship browser updates rather than pretend the contract is stable. Nineteen tools, executed through a queue so two calls cannot fight over one browser. It is also the only widely-used entry here that can send a message and a connection request, which is the line worth thinking hardest about before you wire it to an agent.

{
  "mcpServers": {
    "mcp-server-linkedin": {
      "command": "uvx",
      "args": ["mcp-server-linkedin@latest"],
      "env": { "UV_HTTP_TIMEOUT": "300" }
    }
  }
}

Best for

Experimenting tonight, for free, on a LinkedIn profile you could stand to lose.

You need

uv (or Docker), a LinkedIn account you are willing to automate, and Python 3.12.4+ if you develop against it

Auth

Your own logged-in browser session — manual login, import from a signed-in Chromium, or a stored cookie. No API key, no OAuth.

Price

Free, open source

Worth knowing. Quoting its own README: “LinkedIn’s User Agreement prohibits automated access, and accounts using automated tools can be restricted or banned. Use at your own risk; there is no guarantee of account safety.” Take the messaging tools seriously in that light — an agent in a retry loop here writes to real people, from your name, and there is no undo.

Where stickerdaniel's server maximises surface, this one maximises restraint, and for a research agent that trade is often the right one. Seven tools, none of which can reach another person: there is no send_message, no connection request, nothing that leaves a trace on somebody else's notifications. The profile tool is the interesting part — you choose which sections to scrape (experience, education, contact, interests, honours, languages, posts, recommendations) rather than paying for a full render every time, which keeps both latency and footprint down. Built on FastMCP and Patchright behind a hexagonal architecture, which in practice means the scraping layer is swappable and the code is unusually pleasant to read if you intend to fork it. Job search carries real filters: date posted, job type, experience level, work mode, easy-apply and sort order.

git clone https://github.com/eliasbiondo/linkedin-mcp-server.git
cd linkedin-mcp-server
uv sync
uv run patchright install
uv run linkedin-mcp-server --login
uv run linkedin-mcp-server

Best for

Read-only research where you want the smallest possible blast radius.

You need

Python 3.12+, uv, and a LinkedIn account to log in once

Auth

One browser login via --login; the session is kept in ~/.linkedin-mcp-server

Price

Free, open source

Worth knowing. From its disclaimer: “Scraping LinkedIn may violate their Terms of Service. Use responsibly and at your own risk.” The read-only design genuinely lowers the stakes — the worst case is a restricted account, not a message you did not write going out under your name.

A compact TypeScript server covering the four things a sales-flavoured agent asks for: find a person, read the profile, look at jobs, send a message. It sits on an Axios REST client with what the README calls automatic token management. That phrase is doing a lot of work and the documentation does not unpack it — it never says which token, obtained how, or what happens when it expires. That is not a reason to dismiss the project, but it is a reason to read the source before you hand it credentials, and the README itself says as much. Worth knowing that this is the smallest codebase here, which cuts both ways: less to audit, and less written down.

The README describes capabilities rather than enumerating tool names: profile search and retrieval, job search, direct messaging to connections, and connection statistics.

Best for

Reading, if you want a compact TypeScript reference implementation to learn from or fork.

You need

Node, and LinkedIn credentials whose exact form you should confirm in the source first

Auth

Token-based with automatic token management; the README does not say which token, which is worth resolving before you trust it with credentials

Price

Free, open source

Worth knowing. Its own documentation notes it “handles sensitive LinkedIn authentication credentials” and asks you to review the token-management code yourself before use. Read that literally rather than as boilerplate.

Everything else on this page is either automating a human session or buying scraped data. This one asks LinkedIn for permission and gets it. It uses the official REST API over OAuth 2.0, with two products that LinkedIn grants instantly for personal profiles: Share on LinkedIn (w_member_social) for publishing, commenting and reacting, and Sign In with OpenID Connect for identifying the author. No password leaves your machine, no browser is driven, and nothing here can get your account restricted, because none of it breaks a rule. The price of that is the surface. LinkedIn reserves reading — your own posts, their engagement stats, the feed, search, messages, connections, anybody else's profile — for approved partners, so r_member_social tools are listed but will not work for you. What remains is genuinely useful for one job: an agent that publishes and engages on your behalf. Posts support link previews, images, real clickable hashtags and visibility control.

The last three need r_member_social, which LinkedIn grants only to approved partners — they are present but will not work on a normal developer app.

npx -y github:vlpmedialtd/linkedin-mcp auth

{
  "mcpServers": {
    "linkedin": {
      "command": "npx",
      "args": ["-y", "github:vlpmedialtd/linkedin-mcp"]
    }
  }
}

Best for

An agent that posts and engages on your own profile, entirely within LinkedIn's rules.

You need

Node.js 22+, and a LinkedIn developer app (creating one requires a LinkedIn Page, though posts still land on your personal profile)

Auth

OAuth 2.0, authorization-code flow against LinkedIn's developer platform

Price

Free, open source

Worth knowing. The 60-day token with no refresh is the operational catch nobody mentions until it bites: an always-on posting agent goes silent every two months until a human re-authorises it. Plan for that, or plan to notice.

A TypeScript server built on OAuth 2.0 with token refresh and environment-isolated credentials, aimed squarely at recruiting and lead-generation use cases. The code is the easy part; the hard part is not in this repository. Everything it offers — people search with filters, profile retrieval, job-market data, messaging — sits behind LinkedIn API scopes that LinkedIn does not hand out on request. Whether this server does anything for you depends entirely on what your developer app is approved for, and for most people that approval is the wall. Treat it as a well-built client waiting for permissions rather than as something you install and use tonight.

Documented as capabilities rather than names: advanced people search filtered by location, industry and company; detailed profile retrieval; job-market intelligence; messaging.

Best for

Teams who already hold approved LinkedIn API access and want a clean MCP client for it.

You need

Node.js 16+ and a LinkedIn Developer account with the relevant products approved

Auth

OAuth 2.0 with token refresh; needs your own LinkedIn Developer credentials

Price

Free, open source

Worth knowing. The README states it is “an independent innovation and is not officially affiliated with LinkedIn or Microsoft” and puts GDPR and API-terms compliance on you. The practical caveat is simpler: approval, not code, is what stands between you and these tools working.

The pragmatic answer when you want LinkedIn data and you are not willing to put an account behind it. Bright Data runs the collection on their own proxies and their own identities; you send a URL or a search and get structured JSON back, with bot detection and CAPTCHAs handled upstream as part of the request rather than as your problem. Five of the 69 tools are LinkedIn-specific and they cover the research half well: person profiles with experience and skills, company profiles, job listings, post content and people search. Batch operations take ten items at a time. The free tier is unusually real — 5,000 requests a month, no card — which is enough to build and evaluate something properly before deciding. What you are giving up is any notion of a session: there is no inbox, no messaging, no sending, because there is no account of yours involved. That is the same sentence as the safety guarantee, read from the other side.

Five LinkedIn tools inside a 69-tool set that also spans e-commerce, finance, travel, app stores and browser automation.

# hosted — paste this URL into any MCP client
https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN_HERE

# or run it locally
{
  "mcpServers": {
    "Bright Data": {
      "command": "npx",
      "args": ["@brightdata/mcp"],
      "env": { "API_TOKEN": "<your-api-token-here>" }
    }
  }
}

Best for

Reading LinkedIn data at volume without putting any account at risk.

You need

A Bright Data account and an API token. Nothing else.

Auth

API token from your Bright Data account

Price

5,000 requests/month free, no card; then roughly $1.50 per 1,000 search results

Worth knowing. The safety and the limitation are the same fact: no account of yours is involved, so nothing can be restricted, and nothing can be sent. If the job ends in a message to a human, this is the wrong tool and no amount of budget changes that.

Structurally different from everything else here: it is not a LinkedIn server, it is a server that lets an agent go shopping for one. The agent searches the Apify Store at runtime, discovers Actors — including several LinkedIn scrapers built by third parties — and runs them on Apify's infrastructure, paying per event. That dynamic discovery is genuinely powerful and it is also the catch: the LinkedIn capability is not in this repository and not maintained by Apify, so quality, legality and upkeep vary by whichever Actor your agent happens to pick. The payment options are the most unusual on this page, spanning an ordinary API token, OAuth on the hosted server, and on-chain settlement via x402 on Base.

Only these two are preconfigured. Everything else — LinkedIn scrapers included — is discovered from the Actor Store at runtime, so the tool list is not fixed.

https://mcp.apify.com

Best for

Agents that need to reach many different sites, of which LinkedIn is only one.

You need

An Apify account and an API token, or OAuth on the hosted server

Auth

Apify API token, or OAuth against the hosted server

Price

Pay-per-event, priced per Actor; discovery and docs tools are free

Worth knowing. Judge the Actor, not the server. This project is well maintained; the LinkedIn scraper your agent picks out of the store on a given Tuesday might not be.

Our own entry, so read it with that in mind. The design starts from the thing every open-source server on this page warns you about in its README: the account. Calls run through a sender account connected to your workspace, which is either one you own or one you rent, and never your personal profile — so the tools are not automating you. Underneath the tool boundary sit the rails that a raw API has no reason to have: a per-workspace rate limit on LinkedIn calls, warm-up ramps and weekly caps per sender, working-hours windows, randomised gaps between actions, and a human approval queue on anything irreversible. Every tool reads. Nothing here sends an invite or a message, and that is a boundary rather than a backlog item. The honest trade is that this is a hosted subscription with a prerequisite, not a repository you clone tonight — and for plenty of jobs on this page, the free servers above are the better answer.

claude mcp add --transport http topclozer https://topclozer.com/api/mcp \
  --header "Authorization: Bearer tcz_live_YOUR_KEY"

Best for

Outreach research that still has to be working next month, on an account you can afford to keep.

You need

A workspace, an API key, and a connected LinkedIn account that is not your personal profile

Auth

Workspace API key as a bearer token; no cookie leaves your machine because there is no cookie

Price

From $39/month for the software; LinkedIn accounts are not included

Worth knowing. This is us. If you want a free server you can run tonight and you are fine risking a throwaway profile, one of the open-source entries above is genuinely the better answer, and the comparison table below says so in four cases out of five.

Not a LinkedIn tool, and it is here because profile data on its own rarely finishes the job. A headline tells you someone is VP Engineering at a company you have never heard of; it does not tell you whether that company raised last quarter, shipped something relevant, or is quietly laying people off. Exa is a search index designed to be read by a model rather than a human, and it is the cheapest way to close that gap inside the same conversation. The hosted endpoint works anonymously with rate limits, which means an agent can use it before anybody has signed up for anything.

The first two are on by default; the advanced search adds domain, date and highlight filters, and agent_run performs multi-step enrichment with structured output.

https://mcp.exa.ai/mcp

Best for

The second half of prospect research, after LinkedIn has given you the names.

You need

Nothing to start; an account for higher rate limits

Auth

OAuth in most clients; anonymous with rate limits, or an API key

Price

Free tier with rate limits; higher limits need an account

Worth knowing. It is in this directory because a research agent that only reads LinkedIn produces confident briefings built on one source. That is a worse failure than a missing tool.

The other half of the loop, and the half most LinkedIn MCP setups forget. An agent that researches beautifully and writes its findings into a chat transcript has produced something nobody reads twice; the value only compounds when the result lands in the system your team already opens every morning. This is HubSpot's own hosted server, reached at mcp.hubspot.com over OAuth, with read and write access to standard CRM objects — contacts, deals, engagements — and read-only access to organisational and marketing context. Setup is not trivial: you need to be on HubSpot's new developer platform and create a user-level app scoped to the objects you want.

Read and write on standard CRM objects — contacts, deals, engagements — plus read-only organisational and marketing context. HubSpot documents scopes rather than a fixed tool list.

https://mcp.hubspot.com

Best for

Closing the loop, so what the agent learns survives the conversation.

You need

A HubSpot account on the new developer platform, plus a user-level app with the right read scopes

Auth

OAuth 2.0 via a user-level app on the HubSpot developer platform

Price

Not published on the docs page

Worth knowing. Listed as the destination rather than the source. If your agent has nowhere to write, the rest of this directory produces very good chat logs and nothing else.