LinkedIn MCP Server

A Model Context Protocol server that lets Claude — and any MCP-compatible client — draft, format, analyze, and publish LinkedIn posts straight from a chat.

Stop context-switching between your editor, a "best time to post" blog, and LinkedIn's composer. This server exposes five focused tools so the model can take a raw idea, shape it into a polished post, score it, generate hashtags, and (optionally) publish it to your profile — all without leaving the conversation.

Features

• ✍️ Draft full posts from a topic, tone, and a few key points

• 🎨 Format raw text with clean line breaks, spacing, emojis, and hashtags

• 🏷️ Generate relevant, topic-aware hashtags

• 📊 Analyze a post and get a score plus concrete improvement suggestions

• 🚀 Publish to LinkedIn (or save as a draft) with configurable visibility

• 🧩 Works with Claude Desktop and any other MCP host over stdio

• ⚡ Bundled with esbuild for fast, low-memory builds

Note: Four of the five tools (draft_post, format_post, generate_hashtags, analyze_post) run fully offline and need no credentials. Only create_post talks to the LinkedIn API and requires an access token.

Related MCP server: linkedin-mcp-server

Available Tools

draft_post

Generate a LinkedIn post template from a topic and tone.

format_post

Format raw content with proper LinkedIn structure and hashtags.

generate_hashtags

Generate relevant LinkedIn hashtags for a topic.

analyze_post

Score a LinkedIn post and return improvement suggestions.

create_post

Format and publish a LinkedIn post, or save it as a draft. Requires LinkedIn API credentials.

Prerequisites

• Node.js 18+ and npm (pnpm or yarn work too)

• An MCP-compatible client — e.g. Claude Desktop

• (Only for publishing) A LinkedIn developer app and an access token with the w_member_social scope

Installation

git clone https://github.com/<your-username>/linkedin-mcp-server.git
cd linkedin-mcp-server
npm install
npm run build

The build step bundles src/ into a single executable file at dist/index.js.

Configuration

1. LinkedIn API credentials (for create_post)

1. Create an app in the LinkedIn Developer Portal.

2. Request the Share on LinkedIn product to unlock the w_member_social scope.

3. Complete the OAuth 2.0 flow to obtain an access token.

Create a .env file in the project root:

LINKEDIN_ACCESS_TOKEN=your_access_token_here
# Optional — many setups resolve this from the token via /userinfo.
# Provide it only if your implementation expects it:
LINKEDIN_AUTHOR_URN=urn:li:person:XXXXXXXXXX

⚠️ Never commit your .env file or share your access token. Add .env to .gitignore.

2. Connect to Claude Desktop

Add the server to your Claude Desktop config:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "linkedin": {
      "command": "node",
      "args": ["/absolute/path/to/linkedin-mcp-server/dist/index.js"],
      "env": {
        "LINKEDIN_ACCESS_TOKEN": "your_access_token_here",
        "LINKEDIN_AUTHOR_URN": "urn:li:person:XXXXXXXXXX"
      }
    }
  }
}

Use an absolute path to dist/index.js, then fully restart Claude Desktop. The LinkedIn tools should now appear in the tools menu.

Usage

Once connected, just talk to the model naturally — it will pick the right tool:

• "Draft a LinkedIn post about shipping my first MCP server. Educational tone, aimed at junior devs, and include a CTA to star the repo."

• "Format this draft, add line breaks and a couple of emojis: …"

• "Generate 8 hashtags for a post about AI compliance for fintech."

• "Analyze this post and tell me how to make it stronger: …"

• "Publish this to LinkedIn, connections-only."

Development

Why esbuild?

This project bundles with esbuild rather than tsc. Beyond being dramatically faster, it sidesteps the JavaScript heap out of memory errors that tsc can hit when bundling the MCP SDK and its dependencies. tsc is still used for type-checking only.

Scripts

npm run build      # Bundle src/ -> dist/index.js with esbuild
npm run typecheck  # Type-check with tsc (no emit)
npm start          # Run the built server directly (stdio)

Example esbuild.config.js

import { build } from "esbuild";

build({
  entryPoints: ["src/index.ts"],
  outfile: "dist/index.js",
  bundle: true,
  platform: "node",
  format: "esm",
  target: "node18",
  banner: { js: "#!/usr/bin/env node" },
}).catch(() => process.exit(1));

Project structure

linkedin-mcp-server/
├── src/
│   ├── index.ts            # MCP server entry — stdio transport + tool registration
│   ├── tools/              # One module per tool
│   │   ├── draftPost.ts
│   │   ├── formatPost.ts
│   │   ├── generateHashtags.ts
│   │   ├── analyzePost.ts
│   │   └── createPost.ts
│   └── linkedin/           # LinkedIn API client (auth + posts)
├── dist/                   # Bundled output (esbuild)
├── esbuild.config.js
├── package.json
├── tsconfig.json
└── README.md

Built on top of @modelcontextprotocol/sdk.

Troubleshooting

Tools don't appear in Claude Desktop. Double-check that the path in args is absolute, that npm run build succeeded, and fully quit and reopen Claude Desktop. On macOS, logs live at ~/Library/Logs/Claude/.

JavaScript heap out of memory during build. Use the bundled esbuild build (npm run build). If you reintroduce tsc for emitting, raise the heap limit with NODE_OPTIONS=--max-old-space-size=4096.

401 Unauthorized from create_post. Your access token is missing, expired, or lacks the w_member_social scope. Regenerate it and update your .env / config.

License

MIT

Built with the Model Context Protocol. Contributions and issues welcome.