LinkedIn OpenID MCP

Multi-user Model Context Protocol server that connects ChatGPT / Claude / Cursor to LinkedIn using only LinkedIn’s official Sign In with OpenID Connect API.

Docs site: https://ashishodu2023.github.io/linkedin-mcp/

What it does

Each end-user authorizes their own LinkedIn account (OAuth). No scraping. No messaging automation.

Related MCP server: EnrichB2B MCP Server

Quick start (local)

python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

export BASE_URL="https://YOUR-PUBLIC-URL"          # ngrok / cloud URL
export LINKEDIN_CLIENT_ID="..."
export LINKEDIN_CLIENT_SECRET="..."
export LINKEDIN_REDIRECT_URI="${BASE_URL}/oauth/linkedin/callback"

PYTHONPATH=src uvicorn linkedin_mcp.server:app --reload --port 8080

LinkedIn Developer app

1. Create app at https://www.linkedin.com/developers/apps

2. Products → enable Sign In with LinkedIn using OpenID Connect

3. Auth → Authorized redirect URL: https://YOUR-PUBLIC-URL/oauth/linkedin/callback

ChatGPT

1. Settings → Connectors → Advanced → enable Developer mode

2. Create connector → MCP URL = https://YOUR-PUBLIC-URL/

3. Auth = OAuth (discovery via /.well-known/oauth-authorization-server)

4. Sign in → authorize LinkedIn → ask: Call get_linkedin_profile

Deploy

Any HTTPS host with a public root URL works (Fly.io, Railway, Render, Cloud Run).

# example
export BASE_URL=https://linkedin-mcp.example.com
# set LinkedIn secrets as platform env vars

See Dockerfile and fly.toml / railway.toml in-repo.

Limits (important)

Official LinkedIn OpenID returns lite profile only. It does not grant feed, connections, inbox, or arbitrary profile scraping. Requesting those via unofficial APIs risks account bans.

License

MIT — Ashish Verma