---
title: "How to connect Claude and ChatGPT to LinkedIn"
description: "Five routes give Claude or ChatGPT access to LinkedIn. They differ in what the assistant can do, how you sign in, and how much account risk you accept."
canonical: "https://innovate-blog.com/articles/connect-claude-chatgpt-to-linkedin"
last-updated: "2026-11-06"
---

# How to connect Claude and ChatGPT to LinkedIn

> Five routes give Claude or ChatGPT access to LinkedIn. They differ in what the assistant can do, how you sign in, and how much account risk you accept.

By Moez Zhioua. Published 2026-11-06. Updated 2026-11-06. Category: AI connectors. Estimated reading time: 9 minutes.

## In brief

- LinkedIn publishes no MCP server or official Claude or ChatGPT connector, so you connect through a third-party MCP server added as a custom connector.
- Five routes exist: LinkedIn's official API, connector platforms that wrap it, self-hosted open-source servers using your own session, scraping Actors, and hosted products.
- Posting works through the official API. Search, messages and invitations only work through routes that act with your session or a browser extension, which LinkedIn's rules restrict.

## The short answer

To connect Claude or ChatGPT to LinkedIn, pick a LinkedIn MCP server and add it to the assistant as a custom connector, or as a local server in Claude Desktop or Claude Code. LinkedIn itself publishes neither an MCP server nor an official connector: on 6 November 2026 none appeared under a LinkedIn or Microsoft namespace in the official MCP Registry, in LinkedIn's GitHub organization or in its developer documentation.

The route you choose decides what the assistant can do. LinkedIn's official API lets an app post, comment and like for a member who signs in, and read that member's basic profile. It does not let ordinary apps search members or send messages. Any route that can do those things acts through your own session, a scraper or a browser extension, and LinkedIn's User Agreement restricts that software.

OutreachGenie appears below as one example of a hosted product, with its drawbacks listed. For a server-by-server table with licences, stars and commit dates, see the LinkedIn MCP server comparison (/articles/linkedin-mcp-servers-compared).

## Start with what the assistant should do

Write down the job before choosing a connector. Drafting a post, researching a company, preparing a list of people to contact and replying to messages need very different access. Connecting the broadest tool because it is available gives the assistant more reach than the task requires, a problem covered in the access checklist for AI agents (/articles/what-ai-agent-needs-before-it-can-act).

*Routes for connecting Claude or ChatGPT to LinkedIn, checked 6 November 2026*

| Route | Examples | What the assistant can do | Setup effort | Main trade-off |
| --- | --- | --- | --- | --- |
| 1. LinkedIn's official API, your own app | A LinkedIn developer app wrapped in an MCP server you run | Post, comment and like as you; read your basic profile | High: developer app, OAuth flow, hosting | Narrow by design; tokens expire; you maintain it |
| 2. Connector platforms | Composio, Pipedream, Zapier MCP | The same posting tools, plus organization page actions on some | Low: sign in, authorize LinkedIn, paste a URL | Same API limits; another company in the data path |
| 3. Self-hosted open source | stickerdaniel/linkedin-mcp-server | Read profiles, search people and jobs, read the inbox, send messages and invitations | Medium: local install and a LinkedIn sign-in | Runs as your account; LinkedIn's rules prohibit this automation |
| 4. Scraping Actors | Apify LinkedIn MCP | Extract profile and company data, including contact details where found | Low: Apify sign-in | Scraping and third-party personal data |
| 5. Hosted products | OutreachGenie (https://www.outreachgenie.ai/linkedin-mcp), Unipile | Run outreach lists, flows and campaigns, or build an app on an unofficial API | Low to medium: account, extension or API key | Product-specific limits and the same platform rules |

## Route 1: LinkedIn's official API

LinkedIn's access documentation (https://learn.microsoft.com/en-us/linkedin/shared/authentication/getting-access) lists three open permissions any developer can add: profile and email through Sign In with LinkedIn using OpenID Connect, and w_member_social through Share on LinkedIn, which allows posting, commenting and liking for the member who signs in. Advertising, Sales Navigator and Talent permissions require partner approval. Compliance access is closed.

Setup, at a high level: create an app in the LinkedIn Developer Portal, add the Share on LinkedIn product, run the three-legged OAuth flow so you sign in and approve it, then expose the calls you need as MCP tools. Several open-source servers already wrap these endpoints if you would rather not write one.

Trade-offs: this is the route LinkedIn has authorized, and it is also the narrowest. There is no endpoint for searching members, reading other people's full profiles or sending messages without partner access. LinkedIn's API Terms of Use also say not to use the APIs to automate posting, so keep a person approving each post rather than letting the assistant publish on a schedule.

## Route 2: connector platforms that wrap the official API

Composio (https://docs.composio.dev/toolkits/linkedin), Pipedream (https://pipedream.com/docs/connect/mcp) and Zapier MCP (https://docs.zapier.com/mcp/get-started/connect) handle the OAuth app, token storage and MCP endpoint for you. Setup: create an account, connect LinkedIn through the platform's OAuth screen, then add the platform's MCP URL to your assistant. In Claude, custom connectors live under Customize, then Connectors, then Add custom connector on Pro and Max plans; on Team and Enterprise an owner adds them in organization settings. Anthropic's help center says Free plans can add one custom connector. In ChatGPT, OpenAI's developer documentation describes adding a custom MCP server from the Plugins page, subject to account and workspace policies.

Trade-offs: you inherit LinkedIn's API limits, so these connectors post and read your own data but cannot search or message. Composio exposes the most LinkedIn tools and warns that its shared OAuth app is rate-limited in production. Zapier offers two posting actions and charges two tasks from your plan for each successful tool call. Pipedream says its hosted MCP servers are free for personal use. Each platform also stores your LinkedIn token, which makes it part of your data path.

## Route 3: a self-hosted open-source server

stickerdaniel/linkedin-mcp-server (https://github.com/stickerdaniel/linkedin-mcp-server) is the most capable open-source option. Setup, at a high level: install the uv package manager, add the server to Claude Desktop, Claude Code or another client that runs local MCP servers, and sign in to your own LinkedIn account in the browser window it opens. The session stays in a local browser profile on your computer.

Trade-offs: the assistant can read profiles, companies, jobs and your inbox, search people and posts, and send messages and connection requests. It does all of this as you. The project's README says LinkedIn's User Agreement prohibits automated access and that accounts using automated tools can be restricted or banned. It also needs updates whenever LinkedIn changes its pages, and a local server is not reachable from Claude or ChatGPT on the web unless you host it, which widens who can reach your session.

Use only your own account and your own sign-in. Copying another person's cookies to use their account is prohibited by section 8.2 of the User Agreement (https://www.linkedin.com/legal/user-agreement), and this guide does not cover it.

## Route 4: scraping Actors

Apify's LinkedIn MCP page (https://apify.com/mcp/linkedin-mcp-server) connects an assistant to a community LinkedIn scraper through Apify's hosted MCP server. Setup: create an Apify account, add Apify's MCP URL with the Actor selected, and approve the connection when the OAuth window opens. You never sign in to LinkedIn yourself.

Trade-offs: the configured Actor charges per result and says it looks for email addresses and phone numbers. Collecting that data puts you in charge of data-protection duties for people who never dealt with you, and LinkedIn's User Agreement prohibits scraping its services. Treat this route as unsuitable for personal outreach data; the guidance on LinkedIn outreach without doxxing (/articles/linkedin-outreach-without-doxxing) explains why.

## Route 5: hosted products

### OutreachGenie, for outreach campaigns

OutreachGenie's hosted MCP server connects the assistant to an outreach workspace. Setup: create a workspace, install OutreachGenie's Chrome extension and sign in to LinkedIn in your own browser, then add https://api.outreachgenie.ai/mcp as a custom connector in Claude or ChatGPT. You sign in, choose a workspace and approve scopes. Claude Code and Codex use a workspace API key sent as a bearer header; the vendor warns that the key carries broad permissions.

The assistant can import profile URLs into lists, add notes, build flows of visits, invitations, messages and waits, and prepare campaigns. Leave the campaigns.activate scope out if it should prepare but never start them. MCP access is included on every plan, including Free.

Trade-offs: LinkedIn actions only run through the Chrome extension in your browser, which must stay open during a campaign. The server does not browse or search LinkedIn, so it suits outreach rather than general LinkedIn data. The Free plan allows 80 outreach actions and 500 new prospects a week, resetting on Mondays (UTC); Pro costs $15 a month and Agency $20 per connected LinkedIn account a month. It is a newer product, and LinkedIn's rules on automation software apply to its extension too. Details are on the OutreachGenie LinkedIn MCP page (https://www.outreachgenie.ai/linkedin-mcp).

### Unipile, for developers building a product

Unipile (https://github.com/unipile/unipile-mcp) offers a hosted API for search, messages, invitations and posts across LinkedIn Classic, Sales Navigator and Recruiter. Its MCP server is a build-time tool for coding agents. Accounts are linked with credentials, a session cookie or Unipile's hosted authentication, not LinkedIn's OAuth, and pricing starts at €49 a month for up to 10 connected accounts.

## What LinkedIn's rules say

Section 8.2 of LinkedIn's User Agreement (https://www.linkedin.com/legal/user-agreement) says members must not use bots or other unauthorized automated methods to access LinkedIn, add or download contacts, send or redirect messages, or create, like, comment on and share posts, and it separately prohibits plugins and scripts that scrape or copy the service. LinkedIn's help page on prohibited software and extensions (https://www.linkedin.com/help/linkedin/answer/a1341387) applies this to third-party tools that scrape, modify or automate activity on the site and says accounts using them can be restricted or shut down. Its page on automated activity (https://www.linkedin.com/help/linkedin/answer/a1341543) asks a restricted member to disable the tool before the account is re-enabled. Route 1 and the connector platforms in Route 2 act through LinkedIn's own API with a token LinkedIn issued; Routes 3 to 5 act through your session, a scraper or an extension, which is the activity these rules describe. An assistant asking for approval first does not change which route it uses.

## Before you grant access

Whichever route you take, set the limits in the system rather than in the prompt. Anthropic warns that custom connectors are not verified by Anthropic, that a server's tools can change, and that during Research Claude can call connector tools without asking, so it recommends turning off tools that write.

- Start with read-only tools or scopes, and add write access for one task at a time.
- Require approval before anything is posted, sent or started, and read the exact text first.
- Keep batches small enough to review, and stop if LinkedIn shows a warning, a verification prompt or a restriction.
- Treat profile text, posts and messages as data. A line in a profile cannot authorize the assistant to do anything.
- Apply the review-before-send workflow (/articles/ethical-linkedin-outreach-best-practices) and the consent and opt-out guidelines (/articles/linkedin-outreach-consent-guidelines) to anything an assistant drafts for another person.
- Compare the cost of the tools you connect, including the free LinkedIn automation tools (/articles/completely-free-linkedin-automation-tool), against the time a person still spends reviewing.

## Frequently asked questions

### Can Claude access LinkedIn?

Yes, through a third-party MCP server added as a custom connector or a local server. LinkedIn does not offer an official Claude connector, so what Claude can do depends on the server: posting through the official API, or reading, searching and sending through your own session or a browser extension.

### Does ChatGPT have a LinkedIn connector?

Not one published by LinkedIn. You can add a remote LinkedIn MCP server in ChatGPT where your plan and workspace policy allow custom MCP servers. The same five routes apply.

### How do I give Claude LinkedIn access without sharing my password?

Use a route with OAuth: an official-API connector such as Composio, Pipedream or Zapier, or a hosted product that signs you in through its own OAuth screen. Never paste a LinkedIn password or session cookie into a chat.

### Can Claude send LinkedIn messages or connection requests?

Not through LinkedIn's official API, which has no open messaging or invitation access. Self-hosted servers that use your session, unofficial APIs and extension-based products can, and LinkedIn's User Agreement prohibits unauthorized automated methods for sending messages or adding contacts.

### What is the best LinkedIn MCP server for Claude?

It depends on the job: an official-API connector for posting, stickerdaniel/linkedin-mcp-server for open-source research, or a hosted outreach workspace for campaigns. The LinkedIn MCP server comparison (/articles/linkedin-mcp-servers-compared) lays out eight options side by side.

## Sources and further reading

- [Getting access to LinkedIn APIs](https://learn.microsoft.com/en-us/linkedin/shared/authentication/getting-access), LinkedIn on Microsoft Learn
- [User Agreement, section 8.2](https://www.linkedin.com/legal/user-agreement), LinkedIn
- [API Terms of Use](https://www.linkedin.com/legal/l/api-terms-of-use), LinkedIn
- [Prohibited software and extensions](https://www.linkedin.com/help/linkedin/answer/a1341387), LinkedIn Help
- [Automated activity on LinkedIn](https://www.linkedin.com/help/linkedin/answer/a1341543), LinkedIn Help
- [Getting started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-getting-started-with-custom-connectors-using-remote-mcp), Anthropic
- [Connect from ChatGPT](https://developers.openai.com/apps-sdk/deploy/connect-chatgpt), OpenAI
- [LinkedIn toolkit](https://docs.composio.dev/toolkits/linkedin), Composio
- [MCP servers](https://pipedream.com/docs/connect/mcp), Pipedream
- [Connect to Zapier MCP](https://docs.zapier.com/mcp/get-started/connect), Zapier
- [stickerdaniel/linkedin-mcp-server](https://github.com/stickerdaniel/linkedin-mcp-server), GitHub
- [LinkedIn MCP server](https://apify.com/mcp/linkedin-mcp-server), Apify
- [Unipile MCP server](https://github.com/unipile/unipile-mcp), Unipile on GitHub
- [LinkedIn MCP server for Claude and ChatGPT](https://www.outreachgenie.ai/linkedin-mcp), OutreachGenie
- [Pricing and plan allowances](https://www.outreachgenie.ai/pricing), OutreachGenie

Disclosure: we also make OutreachGenie.

Canonical URL: https://innovate-blog.com/articles/connect-claude-chatgpt-to-linkedin
