Summary
Publishing with the preview step built in.
The assistant writes the post and Unipile does the actual publishing. Because preview is the default, the failure mode of an over-eager model is a draft you reject rather than something already on your feed.
What it is
An MCP server over the Unipile API that publishes to LinkedIn on your behalf. The assistant writes the post or comment; this handles the publishing, with media attachments and company mentions supported, and every publishing action defaulting to preview mode.
What you get
- A post published to LinkedIn from text the assistant wrote (linkedin_publish)
- A comment posted on a LinkedIn post (linkedin_comment)
- Preview by default: publishing actions are a dry run until you confirm them
- Media attachments from local files or URLs, images and video
- Company mentions resolved automatically through Unipile
- Posts auto-liked immediately after publishing
Requirements
Node.js 18+ and a clone with npm install, plus a Unipile account with your LinkedIn connected. The Unipile API key and DSN go in the MCP config rather than a .env file, and the client needs restarting after the edit.