A Unipile account (paid, from 49 EUR a month, 7 day free trial)
Your account linked in Unipile
A scoped Unipile API key, saved in the app's secret store or connector screen
Built from each app's own docs.
Plan: ChatGPT Business, Enterprise or Edu, on the web, set up by an admin or owner. Pro can add servers that only read.
Create a Unipile account. It is paid, from 49 EUR a month for up to 10 accounts, with a 7 day free trial.
Connect your LinkedIn in Unipile, on Unipile's hosted sign in page.
Make a scoped API key in Unipile. Keep it in Unipile and your app's secret store only.
In ChatGPT, open Plugins. Choose +, then Add custom MCP server.
Name it Unipile. Under Connection, enter https://developer.unipile.com/mcp?branch=v2.0
Set up authentication with your Unipile key in the X-API-KEY header. OpenAI's page doesn't list the auth choices, so check a key field shows before you pay for Unipile.
Tick I understand and want to continue. Select Create as a plugin.
In a new chat, type @ and pick Unipile.
Test it: Ask: list my 10 newest LinkedIn chats. Don't send anything.
Never paste an API key into a chat. Use the app's connector screen.
Gemini
What you need:
Gemini CLI
A Unipile account (paid, from 49 EUR a month, 7 day free trial)
Your account linked in Unipile
A scoped Unipile API key, saved in the app's secret store or connector screen
Plan: Gemini CLI on your computer.
Create a Unipile account. It is paid, from 49 EUR a month for up to 10 accounts, with a 7 day free trial.
Connect your LinkedIn in Unipile, on Unipile's hosted sign in page.
Make a scoped API key in Unipile. Keep it in Unipile and your app's secret store only.
In ~/.gemini/settings.json, add a server with "httpUrl": "https://developer.unipile.com/mcp?branch=v2.0" and a header X-API-KEY set to your key. The key goes in this file, never in a chat.
Start Gemini CLI and run /mcp to check it.
Test it: Ask: list my 10 newest LinkedIn chats. Don't send anything.
What it does: Reads one person's profile and posts, then writes a short first message that fits only them.
Use it when: Your LinkedIn messages sound like a copy and paste.
You get: One short message with a real fact about them and one easy question.
Open in
See the prompt
You are my LinkedIn writer.
The person's LinkedIn page: [their profile link]
What I do: [one line]
First, open their profile and read it, plus their last 3 posts.
No connector? Paste instead: [their profile and last 3 posts]
Write one LinkedIn message for this person only.
1. Open with one real fact from their profile or posts that ties to what I do. No praise, no pitch.
2. One line on the real reason I'm writing: the problem I help people like them with.
3. End with one easy question that leaves the timing up to them. No link.
Rules:
- Under 50 words. No flattery.
- Never write one message for a whole list. One message per person, with their name, a real reason, and what we do.
- Never use the word "saw" in any form. Never open with their name and then mine, like "Sam, Jo here."
- Draft only. Don't send.
- Use only what is in my data and what I tell you. Never make up a name, number, date, price, link or promise.
- If something you need is missing, write [check: what is missing] in that spot and keep going.
Last, save these steps as a skill called 'One Person Message' so you do it this way each time I give you a LinkedIn profile.
Connect first: LinkedIn. The steps for each AI are on one page. Then the prompt reads it for you. It ends by saving itself as a skill.
Built from: SMYKM (Show Me You Know Me) Cheat Sheet, Sam McKenna, Section 3: Compose the perfect email
What it does: Reads the people who liked or commented on your LinkedIn posts and builds a warm lead list.
Use it when: You get likes but don't know who to talk to.
You get: A table of warm leads with their comments and a starter line for each.
Open in
See the prompt
You are my sales researcher, with access to Apify.
Goal: turn these LinkedIn posts into a list of warm leads.
Post links: [paste 2 or 3 links]
My job titles: [title 1], [title 2]
Do this:
1. Use an Apify LinkedIn post reactions and comments scraper on each link.
2. Combine the lists and remove repeats. If someone shows up on more than one post, note how many.
3. Keep only people whose job title matches mine, or is close.
4. For anyone who commented, copy what they said.
5. For each commenter, write one line I could use to start a talk, built only on their comment. No pitch.
Rules:
- Use only what is in my data and what I tell you. Never make up a name, number, date, price, link or promise.
- If something you need is missing, write [check: what is missing] in that spot and keep going.
- Ask me before you send, post, buy or delete anything.
No connector? Paste instead: the list of names and comments under each post.
Give me a table: name, title, company, LinkedIn link, liked or commented, how many posts, their comment, and the starter line.
Put people who commented on more than one post at the top, then people who commented, then likes. The more they already know the topic, the warmer they are.
Last, save these steps as a skill called 'Warm Leads From Posts' so you do it this way each time I ask.
Connect first: LinkedIn. The steps for each AI are on one page. Then the prompt reads it for you. It ends by saving itself as a skill.
Built from: Fanatical Prospecting, Jeb Blount, Chapter 12: The Law of Familiarity
What it does: Reads their last 5 posts and writes a short message that gives a free, useful thing first.
Use it when: Your first messages get ignored or feel pushy.
You get: A 3-line message with one free gift and one question about their problem.
Open in
See the prompt
You are my business coach.
The person's LinkedIn page: [their profile link]
What I'm good at: [one line]
First, open their profile and read their last 5 posts.
No connector? Paste instead: [their last 5 LinkedIn posts]
Do this:
1. Tell me what they care about most right now.
2. Give me 3 small, free things I could make just for them in under an hour, like a short list, a page or a fix.
3. Pick the one they'd value most and remember. Tell me why.
4. Write a 3-line message that gives it with no strings, ending with one question about their problem.
Rules:
- No flattery. Be specific.
- Use only what is in my data and what I tell you. Never make up a name, number, date, price, link or promise.
- If something you need is missing, write [check: what is missing] in that spot and keep going.
Last, save these steps as a skill called 'Give First' so you do it this way each time I give you a LinkedIn profile.
Connect first: LinkedIn. The steps for each AI are on one page. Then the prompt reads it for you. It ends by saving itself as a skill.
Built from: Influence, Robert Cialdini, Chapter 2: Reciprocation
What it does: Reads your old posts and notes, then writes one short post that sounds like you.
Use it when: You want to post but the blank page stops you.
You get: One post under 150 words, in your words, with a closing question.
Open in
See the prompt
You are my ghostwriter.
My LinkedIn: [profile link]
This week's notes: [paste]
First, read my last 10 posts to learn how I sound.
No connector? Paste instead: [2 old posts]
Write one LinkedIn post in my voice.
1. Pick the most useful thing in my notes for a stranger.
2. Start with the result. Say "how I", not "how to".
3. Short lines. One idea. Give what the start promised.
4. End with a question or a takeaway.
Rules:
- No hashtags, no emojis, no big words.
- Under 150 words.
- Only use facts from my notes.
Last, save these steps as a skill called 'Post In My Voice' so you do it this way each time I ask.
Connect first: LinkedIn. The steps for each AI are on one page. Then the prompt reads it for you. It ends by saving itself as a skill.
Built from: $100M Leads, Alex Hormozi, Section III: Get Leads, Post Free Content
What it does: Checks each person on your list against their LinkedIn profile and marks them OK, LEFT or CHECK.
Use it when: You're about to call a list you have not checked in a while.
You get: Counts for OK, LEFT and CHECK, a table of OK people, and notes for each LEFT record.
Open in
See the prompt
You are my data checker, with access to Apify.
Goal: make sure everyone on my list still works at the company I have. Calling the wrong person is the most costly waste of my time.
My list: [the Smart View or list name in my CRM]
Do this:
1. Read that list in my CRM: name, company and LinkedIn link for each person. Then use an Apify LinkedIn profile scraper on every link.
2. Compare their company today with the company on my list.
3. Mark each person OK (same company), LEFT (different company) or CHECK (can't tell).
4. For each LEFT person, write the note I should save on their record: where they work now and the date you checked.
Rules:
- A profile that won't load is unconfirmed, never left. Mark it CHECK.
- If the scraped company matches the company on my list, mark the person OK. Only mark CHECK if the profile did not load or the company is missing.
- The date you checked must come from a date I give you or the current date. If you do not have one, write "not supplied". Never guess a date.
- This only catches people who updated their LinkedIn. Say so when you give me the counts.
- Anyone who LEFT never gets called. Their company goes back on the list so we find the person in that seat now.
- Use only what is in my data and what I tell you. Never make up a name, number, date, price, link or promise.
- If something you need is missing, write [check: what is missing] in that spot and keep going.
No connector? Paste instead: name, company and LinkedIn link for each person.
Give me the count of OK, LEFT and CHECK, then a table of the OK people only.
Last, for each LEFT company, tell me what job title to search for to find who took over their seat. If a LEFT person moved to a company that fits my buyer, list them too. They may know me already.
Last, save these steps as a skill called 'Still In The Seat' so you do it this way each time I ask.
Connect first: your CRM and LinkedIn. The steps for each AI are on one page. Then the prompt reads it for you. It ends by saving itself as a skill.
Built from: Fanatical Prospecting, Jeb Blount, Chapter 11: Own Your Database
What it does: Reads your client list, picks 15 people with the same problem, and drafts the invite, follow up, and reminder.
Use it when: You want a small room of people who share one problem.
You get: A guest list with reasons, a plain event name, and 3 short invite lines.
Open in
See the prompt
You are my event planner.
Who I want in the room: [job or problem, for example owners of small cleaning companies]
My list: my clients and leads in my CRM
No connector? Paste instead: [your client list or LinkedIn connections export]
Where: [city]
Do this:
1. Read my list. Pick the 15 people on it who fit best. They should share the same work and the same problem. Give one short reason for each.
2. Plan a simple evening where people trade notes on one problem. No sales pitch. Give it a plain name that says who it's for.
3. Write what I say when I call to invite someone. 3 short lines: who it's for, where and when.
4. Write a short text or email with the details, to send after they say yes. Ask them for one question they'd want the group to answer.
5. Write a reminder for the day before.
Rules:
- Friendly and short. Plain words.
- Only say who's coming once they've said yes. Never claim a guest who hasn't.
- Drafts only. I invite each person myself.
- Use only what is in my data and what I tell you. Never make up a name, number, date, price, link or promise.
- If something you need is missing, write [check: what is missing] in that spot and keep going.
Last, save these steps as a skill called 'Plan A Meetup' so you do it this way each time I ask to plan a meetup.
Connect first: your CRM and LinkedIn. The steps for each AI are on one page. Then the prompt reads it for you. It ends by saving itself as a skill.
Built from: Influence, Robert Cialdini, Chapter 8: Unity
Simple AI
+2LinkedInFirst message
