linkedin-mcp

Publish and schedule posts on your own LinkedIn — from an AI assistant, via the official API.

An MCP server that lets Claude (or any MCP client) write a post and publish or schedule it on your LinkedIn profile through LinkedIn's official Posts API. Plus a content-calendar helper so you grow the profile the sustainable way.

✅ What it does — and what it deliberately doesn't

Does: posts as you (immediately or scheduled), lists your queue and history, and gives you proven post structures + a weekly cadence.

Does NOT: auto-connect, auto-like, auto-comment, mass-message, scrape, or run engagement bots. Those violate LinkedIn's User Agreement and get accounts permanently banned. There is no official API for them — only grey scraping — and it's not worth risking your account. Profiles grow from consistent, valuable content, not automation. That's what this tool supports.

Related MCP server: LinkedIn - Full Tools MCP (via Reepl)

Tools

Setup

1. Create a LinkedIn app (once, ~10 min)

1. Go to developer.linkedin.com → My apps → Create app.

2. Fill in name, associate a LinkedIn Page (create a simple one if you don't have it — required by LinkedIn), upload a logo.

3. Open your app → Auth tab:

   • Copy the Client ID and Client Secret.

   • Under OAuth 2.0 redirect URLs, add exactly: http://localhost:8710/callback

4. Open the Products tab and request:

   • Sign In with LinkedIn using OpenID Connect (gives openid profile)

   • Share on LinkedIn (gives w_member_social — the posting scope)

   These are self-serve for personal use and usually approve instantly. If a scope isn't yet granted, the auth step below will tell you which one is missing.

2. Install

cd linkedin-mcp
python3 -m venv .venv && source .venv/bin/activate
pip install -e .

3. Authorize the tool (once)

Put your credentials in the environment (or ~/.linkedin-mcp/token.json):

export LINKEDIN_CLIENT_ID=xxxxxxxx
export LINKEDIN_CLIENT_SECRET=xxxxxxxx
export LINKEDIN_REDIRECT_URI=http://localhost:8710/callback

python -m linkedin_mcp.auth

A browser opens LinkedIn's consent page → approve → the token is saved to ~/.linkedin-mcp/token.json (chmod 600). Done.

4. Connect to Claude

Add to your MCP client config (e.g. Claude Desktop claude_desktop_config.json):

{
  "mcpServers": {
    "linkedin": {
      "command": "python",
      "args": ["-m", "linkedin_mcp.server"],
      "env": {
        "LINKEDIN_CLIENT_ID": "xxxxxxxx",
        "LINKEDIN_CLIENT_SECRET": "xxxxxxxx"
      }
    }
  }
}

Now you can say: "Write a build-in-public post about my AST taint scanner and publish it" — the assistant uses post_brief to shape it, then publish_now.

5. Auto-posting on a schedule

schedule_post queues posts; a small scheduler publishes the due ones. Run it from cron so posts go out even when you're offline:

*/15 * * * * cd /path/to/linkedin-mcp && .venv/bin/python -m linkedin_mcp.scheduler >> ~/.linkedin-mcp/scheduler.log 2>&1

How growth actually works here

Ask the assistant for content_calendar() — it returns a Mon/Wed/Fri cadence and rules (hook first, be specific, one idea, real numbers, end with a question). Then generate posts with post_brief and schedule a week at a time. Consistency

• substance is what moves a profile — and it can't be taken away by a ban.

Security & privacy

• Token and posts are stored locally under ~/.linkedin-mcp/ (chmod 600), never in the repo.

• The tool only calls the official LinkedIn API and only acts as you.

• Access tokens expire (~60 days); the client auto-refreshes when a refresh token is present, otherwise re-run python -m linkedin_mcp.auth.

License

MIT © nadirzhon — for use on your own LinkedIn account, within LinkedIn's terms.