Claude for LinkedIn: write, check and post from Claude

1. Plain Claude chat, with a voice file and your own material

Most AI posts sound generic because the model is guessing twice. It does not know how you write, and it does not know what happened in your week. Fix both inputs and plain Claude gets much better. You need no tools for this, only two prompts.

Prompt 1: build your voice file (once)

Pick three posts you wrote yourself and were happy with. Paste them into this prompt:

I want you to write LinkedIn posts that sound like me. Below are three posts I wrote myself. Study them and write a one-page voice file I can reuse in every future chat. Cover:
- who I write for, and what they care about
- the positions I hold and would argue for
- how I open a post, and how I end one
- sentence length, rhythm and formatting habits
- words and phrases I use, and ones I never use
For each point, quote a short example from my posts. Do not write a new post yet.

Post 1:
[paste]

Post 2:
[paste]

Post 3:
[paste]

Save the result. In Claude, the easiest home for it is a Project: put the voice file in the project's instructions or knowledge, and every chat in that project starts from it.

Prompt 2: turn your own material into a post

Give Claude something real: a voice-note transcript, meeting notes, a customer email, a Slack thread, a result you shipped. Then:

Use my voice file. Here is raw material from my week:
[paste the transcript, notes, email or thread]

My audience: [who you want to reach]

1. Find the one idea in this material my audience would save or send to a colleague. Tell me why in one line.
2. Write a LinkedIn post about it, 800 to 2,500 characters. The first two lines must work on their own, because LinkedIn cuts the post at "see more" after roughly 140 to 210 characters on mobile.
3. Use only facts from my material. If you need a number, a name or an example I did not give you, ask me instead of inventing one.
4. Then critique the draft like a tough editor: name the three weakest lines, say why, and rewrite them.

The last two steps do the most work. Telling Claude to ask instead of invent stops made-up numbers, and the self-critique catches generic lines before your readers do. You can also paste any draft into the free AI-sounding post checker.

What this way cannot do: Claude does not see your LinkedIn. It cannot post or schedule, and it does not know how past posts did unless you paste the numbers in. Outside a Project, it forgets your voice file between chats.

2. Free LinkedIn skills for Claude

A skill is a folder with a SKILL.md file that teaches Claude a repeatable job. Claude loads it when a task calls for it and runs it on its own model, so there are no extra keys or accounts. Liftli publishes 31 free LinkedIn agent skills, MIT-licensed, on GitHub: post and hook writers, a comment writer, a profile audit, an AI-tell checker, a fold check and more.

Install in one line

npx skills add liftli-ai/linkedin-agent-skills

That works in Claude Code, Codex, Cursor and other skills-compatible agents. In Claude Code you can also add them as a plugin: /plugin marketplace add liftli-ai/linkedin-agent-skills, then /plugin install linkedin-agent-skills@liftli.

In the Claude apps (web and desktop), you upload skills instead. Download a skill folder from the GitHub repo, zip it, and add it under Customize → Skills. Code execution has to be on, under Settings → Capabilities. Anthropic's help article has the current steps.

Where to start

• linkedin-voice builds your voice file from three of your posts and four short questions. Every writing skill reads it before it writes.
• linkedin-agent is the front door. Ask "help me with my LinkedIn" and it routes you to the right skill.
• ai-sounding-post-checker flags the tells that make a draft read as AI, and fixes them by adding specifics.
• linkedin-hook-generator writes 8 openings, each on a different pattern, sized for the "see more" fold.

What skills cannot do: they write and check, but they never post. The linkedin-publish skill hands you a copy-ready post, or publishes through Liftli once you connect it. For a deeper walkthrough, see how to give Claude a LinkedIn skill.

3. An MCP connector that can publish: Liftli

MCP (Model Context Protocol) lets Claude use outside tools. A LinkedIn MCP connector is what turns Claude from a writer into something that can put the post on your profile. Liftli is one: a remote LinkedIn MCP server that publishes and schedules through LinkedIn's official API, and only after you approve.

Connect it

1. Add the URL. In Claude (web, desktop or mobile): Customize → Connectors → Add custom connector, then paste https://mcp.liftli.ai/mcp. In Claude Code: claude mcp add --scope user --transport http liftli https://mcp.liftli.ai/mcp
2. Say "connect my LinkedIn". You approve on LinkedIn's own sign-in page. Liftli never sees your LinkedIn password.
3. Ask for what you want. For example: "Turn this voice note into a LinkedIn post and schedule it for Tuesday at 8:30."

What it adds on top of ways 1 and 2: your voice is extracted from your own posts and remembered across chats and across AI clients. Every post goes through a plan, draft, critique and revise loop. Voice notes, call transcripts and chats become post ideas. And the finished post is published or scheduled for you after one tap. The same URL works in ChatGPT, Codex and Grok.

To see how Liftli compares with the other LinkedIn MCP servers, including the ones that drive your browser session, read the LinkedIn MCP comparison.

What Claude cannot do safely on LinkedIn

Claude can now control a browser, through Claude in Chrome or computer use, and some MCP servers reuse your LinkedIn login cookies. It is tempting to let one of them like, comment, connect or post for you. Don't. LinkedIn's help center says it does not permit third-party software or browser extensions that scrape or automate activity on the site, and that members who use them risk having their accounts restricted or shut down.

• No browser bots. Do not let any AI click around LinkedIn for you: no auto-likes, auto-comments, connection-request runs or profile scraping.
• No posting without a connector. Plain Claude cannot reach LinkedIn. Without a connector that uses the official API, you copy the draft and post it yourself. That is fine.
• No guessing at your results. Claude does not see your LinkedIn analytics. Paste the numbers in if you want it to learn from them.
• No invented facts. Claude fills gaps with plausible numbers and stories unless you tell it not to. Prompt 2 above does.

The safe pattern is the same in all three ways: your material in, your voice out, and a human yes before anything goes live.