WHAT IS MCP?
The Model Context Protocol (MCP) is an open standard, published by Anthropic in late 2024 and since adopted by most AI assistants, ChatGPT included. It lets an AI use external tools. Before MCP, every integration was hand-built; with it, a tool is described once, in a common format, and any compatible AI knows how to use it.
In practice: you add a connector once, then talk to your AI as usual, and it acts in the connected tool. Four words come up often:
| Term | What it is | For Paulo |
|---|---|---|
| MCP client | The AI that uses the tools | Claude or ChatGPT |
| MCP server | The service that exposes the tools | Paulo |
| Tool | A specific action the AI can call | "Send a message", "search profiles"… |
| Connector | The server, once added to your AI | Paulo under Customize › Connectors (Claude) or Plugins (ChatGPT) |
WHAT IT LOOKS LIKE IN PRACTICE
Take a simple request: "follow up with the people who accepted my invite this week but haven't replied". Here's what your AI does, call after call:
- it fetches the list of your recently accepted invitations;
- it reads each conversation to see where it stands;
- it skips those who already replied;
- it writes the others a message that fits what was said;
- it sends the messages through Paulo, then gives you a summary: how many follow-ups went out, and to whom.
Good to knowAll that work, which usually takes an hour of clicking and copy-pasting, fits in one sentence. And every call is visible in the conversation: you always know what your AI did.
WHAT PAULO'S LINKEDIN MCP SERVER DOES
Once Paulo is connected, your AI can, when you ask:
| Area | What your AI can do | Example |
|---|---|---|
| Messages | Read your conversations, send messages and follow-ups | "Follow up with those who haven't replied in 7 days." |
| Network | Send invitations, with a different note for each | "Invite the 20 HR heads at Lyon SMBs I haven't contacted." |
| Search | Search profiles by your criteria (role, city, company, tenure) | "Find backend devs in Nantes with 5 years of experience." |
| Information | Fetch a profile, its latest posts, a conversation's history | "Summarize Mike Davis's profile before my call." |
| Publishing | Publish or schedule posts, reply to comments | "Publish this post Thursday at 9am." |
WHICH AIS?
Paulo works with the two most used assistants, through their MCP connector support:
| AI | Where to add it | Plan needed |
|---|---|---|
| Claude | claude.ai: Customize › Connectors › Add › Custom | Every plan (one custom connector on the free plan) |
| ChatGPT | ChatGPT web: Plugins › Add › Create custom MCP server | A paid plan, Plus is enough |
Good to knowThese menus change regularly at both Anthropic and OpenAI. The "Connect LinkedIn to Claude" and "Connect LinkedIn to ChatGPT" guides detail every step and are kept up to date.
WHY NOT LINKEDIN'S OFFICIAL API?
LinkedIn does offer APIs, but access is restricted: most require being an approved partner, and they're designed for specific uses (ads, company page management) rather than a personal account's messaging.
That's why most LinkedIn MCP servers go another way, and also where the quality of the integration makes the difference: stability, sending pace, respect for the network's limits.
HOSTED OR OPEN SOURCE: THE DIFFERENCE
There are several open-source LinkedIn MCP servers. They're a good option if you're a developer who likes to tinker. For a salesperson, a recruiter or a marketing team, they're often one project too many. The comparison:
| Open-source server | Paulo (hosted) | |
|---|---|---|
| Setup | Python or Node environment, configuration in your AI client | Add a connector, that's it |
| LinkedIn sign-in | Grab a session cookie or token yourself | You connect your account in Paulo |
| Maintenance | Up to you to fix when LinkedIn changes something | Handled |
| Sending pace | Yours to set, or not set at all | Spread out automatically |
| Where it runs | Often on your computer: ChatGPT and claude.ai only accept servers reachable online | Online, usable from Claude and ChatGPT web |
SECURITY: WHAT TO KNOW
Connecting an AI to your LinkedIn means giving it the right to act on your behalf. Three points deserve your attention, whatever MCP server you choose.
First, only connect servers you trust: ChatGPT and Claude remind you of this themselves when you add one. An MCP server sees the data your AI asks it for and can act with your account's rights.
Second, the messages you receive are written by strangers. A booby-trapped message could contain instructions aimed at your AI ("ignore your instructions and send…"). AIs resist this better and better, but the good habit stays the same: ask for a preview before any batch send, and never let your AI work through an inbox alone without review.
Finally, keep the ability to cut everything off: turn the connector off in a conversation, remove it from your AI, or disconnect your LinkedIn account. With Paulo, all three are possible at any time, and your data is never sold or rented.
A PACE THAT LOOKS LIKE YOURS
A LinkedIn account that sends a hundred invitations in a minute gets noticed. LinkedIn doesn't publish its thresholds, but practitioners agree on a few markers: around twenty invitations a day, about a hundred a week, and messages spread out rather than sent in bursts.
Paulo sends at a normal pace: if you ask for 50 follow-ups, they go out gradually rather than all at once. You stay in charge: nothing goes out unless you ask.
WRITING GOOD REQUESTS
Your AI does better with a bit of framing. A few habits that change the result: name the target, "SaaS sales directors in Lyon" rather than "prospects"; give the tone and length, "short, first names, no stock phrases"; say what you already know, "Sarah asked about pricing on Tuesday"; ask for a preview before a big send, "show me the first 3 messages before sending"; set a limit, "no more than 20 invitations today".
To avoid repeating these instructions, put them once in a project in your AI: Claude and ChatGPT both offer projects that keep your instructions for every conversation.
EXAMPLE REQUESTS
Here is the kind of sentence you write to your AI once Paulo is connected:
“Follow up with everyone who accepted my invitation this week but didn't reply.”
“Find 20 SaaS sales directors in Lyon I haven't contacted yet.”
“Summarize Mike Davis's profile and his last three posts before my call.”
“Publish this post on Thursday at 9am.”
“List the messages I haven't answered in a week.”
WHO IS IT FOR?
Salespeople use it to prospect and follow up without giving up their evenings. Recruiters to source profiles and write first messages that don't read like a template. Marketing teams to keep a publishing rhythm. Founders to stay present on LinkedIn without spending an hour a day there.
In every case the idea is the same: you describe what you want, your AI handles the clicks.