LinkedIn MCP Server

Tests

A production-ready Model Context Protocol (MCP) server connecting AI agents (Google Gemini Spark, OpenAI Codex, Claude Desktop, Cursor, Windsurf) to the official LinkedIn REST API.

Enables AI agents to autonomously draft and publish text updates, share rich link previews, upload images, create interactive polls, manage existing posts, and engage with reactions and threaded comments.

https://github.com/user-attachments/assets/cda965be-d6a2-4a37-8d3c-29a68a9959e1

Features

Dual-Mode MCP Transport:
- Server-Sent Events (SSE) via GET /mcp and POST /messages for clients like Google Gemini Spark and Claude.
- Streamable HTTP via POST /mcp for modern command-line and agent runtimes like OpenAI Codex (rmcp).
12 Comprehensive Tools:
- Content Creation: Text updates, article link preview cards, binary image uploads (3-step pipeline), and multi-option polls.
- Content Management: Delete posts by URN, fetch engagement statistics (likes, comments, shares), and verify account profile connections.
- Social Engagement: Add reactions (LIKE, CELEBRATE, LOVE, INSIGHTFUL, FUNNY, EMPATHY), delete reactions, add comments, reply in threads, and remove comments.
Robust Validation: Powered by Zod schemas for all tool input parameters.
Enterprise-Ready: Automated unit and integration test suite with 100% endpoint coverage.

Tool Reference

Tool	Category	Description	Parameters
`create_text_post`	Content	Publish a standard text post to your LinkedIn feed	`content` (string, required)
`create_link_post`	Content	Publish a post with a clickable preview card	`content` (string), `url` (string, URL), `title` (string)
`create_image_post`	Content	Download an image from a URL, upload it to LinkedIn, and publish	`content` (string), `image_url` (string, URL)
`create_poll_post`	Content	Create a poll with customizable duration	`question` (string), `options` (string array, 2-4 items), `duration` (`ONE_DAY`, `THREE_DAYS`, `ONE_WEEK`, `TWO_WEEKS`)
`delete_post`	Management	Delete a post using its Post URN	`post_urn` (string, e.g. `urn:li:ugcPost:...`)
`get_user_profile`	Management	Retrieve authenticated profile details (name, member ID)	None
`get_post_stats`	Management	Fetch engagement metrics (likes, comments, shares)	`post_urn` (string)
`react_to_post`	Engagement	Add a reaction to a post	`post_urn` (string), `reaction_type` (`LIKE`, `CELEBRATE`, `LOVE`, `INSIGHTFUL`, `FUNNY`, `EMPATHY`)
`remove_reaction`	Engagement	Remove your reaction from a post	`post_urn` (string)
`comment_on_post`	Engagement	Add a top-level comment on a post	`post_urn` (string), `comment_text` (string)
`delete_comment`	Engagement	Delete a comment you made	`post_urn` (string), `comment_id` (string)
`reply_to_comment`	Engagement	Post a threaded reply to an existing comment	`post_urn` (string), `parent_comment_urn` (string), `reply_text` (string)

Setup & Authentication

1. LinkedIn Developer Portal Setup

Create a LinkedIn application in the LinkedIn Developer Portal.
Under the Products tab, request access to:
- Share on LinkedIn (provides w_member_social permission)
- Sign In with LinkedIn using OpenID Connect (provides openid and profile permissions)
In the Auth tab, add a redirect URL (e.g. https://httpbin.org/get or your domain).

2. Generate a User Access Token

Construct the authorization URL in your browser:

https://www.linkedin.com/oauth/v2/authorization?response_type=code&client_id=YOUR_CLIENT_ID&redirect_uri=YOUR_REDIRECT_URI&scope=openid%20profile%20w_member_social

Authorize the application and copy the code query parameter from the redirection URL.

Exchange the code for a 60-day access token:

curl -X POST https://www.linkedin.com/oauth/v2/accessToken \
  -d "grant_type=authorization_code" \
  -d "code=YOUR_AUTHORIZATION_CODE" \
  -d "client_id=YOUR_CLIENT_ID" \
  -d "client_secret=YOUR_CLIENT_SECRET" \
  -d "redirect_uri=YOUR_REDIRECT_URI"

Fetch your member ID:
```
curl -H "Authorization: Bearer YOUR_ACCESS_TOKEN" https://api.linkedin.com/v2/userinfo
```
The sub field is your member identifier (e.g., urn:li:person:<sub_id>).

Configuration

Copy the sample environment file:

cp .env.example .env

Configure your .env variables:

LINKEDIN_ACCESS_TOKEN=your_oauth2_access_token
LINKEDIN_AUTHOR_URN=urn:li:person:your_member_sub_id
PORT=3000

Running Locally

# Install dependencies
npm install

# Run unit and integration tests
npm test

# Start the server
npm start

# Start in development mode (with live reload)
npm run dev

Client Integration

OpenAI Codex CLI

Add the server using the Codex CLI:

codex mcp add linkedin --url https://your-server-domain.com/mcp

Or add it directly to ~/.codex/config.toml:

[mcp_servers.linkedin]
url = "https://your-server-domain.com/mcp"

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "linkedin": {
      "command": "node",
      "args": ["/path/to/linkedin-mcp-server/server.js"],
      "env": {
        "LINKEDIN_ACCESS_TOKEN": "your_access_token",
        "LINKEDIN_AUTHOR_URN": "urn:li:person:your_sub_id"
      }
    }
  }
}

Google Gemini Spark

In your Gemini Spark custom applications settings, enter your public SSE endpoint URL:

https://your-server-domain.com/mcp

Deployment

Deploy to Railway

Install the Railway CLI: npm i -g @railway/cli
Link or create a project:
```
railway init
railway up
```

Set environment variables:

railway variable set LINKEDIN_ACCESS_TOKEN="your_token" LINKEDIN_AUTHOR_URN="urn:li:person:your_sub_id"

Generate a public domain:
```
railway domain
```

Development & Testing

The test suite includes 62 automated unit and integration tests covering the LinkedIn API client, all 12 tool registrations and execution handlers, and SSE / Streamable HTTP transports.

# Run tests once
npm test

# Run tests in watch mode
npm run test:watch

License

This project is licensed under the MIT License.

LinkedIn MCP Server