hashan-silva/linkedin-mcp-server

LinkedIn MCP Server (Python, stdio)

Minimal Model Context Protocol (MCP) server that speaks JSON-over-stdio for use with Codex CLI. It wraps LinkedIn REST endpoints so an agent can fetch your profile.

You will need a valid LinkedIn OAuth access token with the appropriate scopes. Tokens are not handled here; provide them via env vars.

Features

• Get profile
• Create post
• Create article post

Project layout

src/
  linkedin_client.py    # HTTP wrapper around LinkedIn REST
  mcp_server.py         # JSON-RPC/stdio MCP server exposing tools
requirements.txt        # Python deps (requests)
Dockerfile

Run locally (stdio)

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
export LINKEDIN_ACCESS_TOKEN="<your-token>"
# optional: override API base
# export LINKEDIN_BASE_URL="https://api.linkedin.com"
python -m src.mcp_server

The server stays attached to stdio. Configure Codex CLI to use the stdio command (see below).

Built-in OAuth helper (fetch token on start)

If you have a LinkedIn app client ID/secret and want the server to launch a local redirect catcher:

export LINKEDIN_CLIENT_ID="your-client-id"
export LINKEDIN_CLIENT_SECRET="your-client-secret"
export LINKEDIN_REDIRECT_URI="http://127.0.0.1:8765/callback"  # must match your app config
# optional: scopes (space-separated)
# export LINKEDIN_SCOPE="r_liteprofile w_member_social"
python -m src.mcp_server --auth

It will print an authorization URL; open it in a browser, approve, and the server will start with the fetched access token. You can still provide LINKEDIN_ACCESS_TOKEN directly if you already have one.

Docker

Build and run:

docker build -t linkedin-mcp-server .
docker run --rm -e LINKEDIN_ACCESS_TOKEN="$LINKEDIN_ACCESS_TOKEN" linkedin-mcp-server

Run with an alternate base URL (e.g., sandbox or proxy):

docker run --rm -e LINKEDIN_ACCESS_TOKEN=... -e LINKEDIN_BASE_URL=https://api.linkedin.com linkedin-mcp-server

Public image guidance: Do not bake secrets into the image. Pass creds at runtime via -e or a bind-mounted .env:

docker run --rm --env-file .env linkedin-mcp-server

An example env file is provided in .env.example. Never commit your filled-in .env.

OAuth in Docker (fetch token on start):

docker run --rm \
  -p 8765:8765 \  # expose redirect port
  -e LINKEDIN_CLIENT_ID=... \
  -e LINKEDIN_CLIENT_SECRET=... \
  -e LINKEDIN_REDIRECT_URI=http://127.0.0.1:8765/callback \
  linkedin-mcp-server python -m src.mcp_server --auth

Open the printed URL, approve, and the container will start the MCP server with the token.

Codex CLI config (stdio server)

Add to your ~/.codex/config.toml (adjust paths as needed):

[mcp_servers.linkedin]
command = "docker"
args = ["run", "--rm","-i","-e","LINKEDIN_ACCESS_TOKEN", "hashan0314/linkedin-mcp-server"]
env = {"LINKEDIN_ACCESS_TOKEN"="<your_token>"}
startup_timeout_ms = 30000

Ensure the working directory is this repo (or set cwd if supported by your CLI).

Exposed tools

• get_profile – fetch current profile
• create_post – create a LinkedIn post
• create_article_post – create a LinkedIn article post with a link

Notes

• This is a lightweight JSON-RPC loop for MCP stdio. Validation is minimal; LinkedIn API errors are returned to the caller.
• Be mindful of LinkedIn API rate limits and scopes (e.g., r_liteprofile, etc.).
• No persistence beyond the LinkedIn API itself.

Backlog / Future Work

• Public distribution without exposing secrets: deliver a brokered-token service (you host LinkedIn client secret and issue short-lived access tokens) or a managed MCP endpoint you run. Avoid embedding client secrets in public images or repos.