Home / Blog / LinkedIn Ads With Claude

How to Connect LinkedIn Ads to Claude or ChatGPT

Quick Answer
You connect LinkedIn Ads to Claude or ChatGPT through an MCP server, which takes about five minutes: sign up, authorise your ad accounts over OAuth, paste the server URL into your assistant's connector settings, and ask a question. The connecting is the easy part. What determines whether the answers are any good is which server you use and what you ask it — because most connectors pass LinkedIn's raw clicks field straight through, and that field counts likes, comments and profile clicks as if they were landing page visits. This guide covers the setup, the first five prompts worth running, and how to check the numbers you get back are real.

What You Need First

Three things:

  • Admin or campaign manager access to the LinkedIn ad account you want to connect. Viewer access is not enough to authorise.
  • An MCP-compatible assistant. Claude and ChatGPT both support MCP connectors today.
  • An MCP server. This is the piece that sits between your ad account and the assistant. There are at least eight for LinkedIn Ads, and they are not equivalent — we compare them in the full LinkedIn Ads MCP guide.

You do not need engineering help, a developer app, or anything installed locally. If a setup guide tells you to run terminal commands, that is a locally-hosted server. Remote servers are simpler and there is no reason to prefer local for this.

Connecting It, Step by Step

1
Sign up
Create an account at try.kiin.co
2
Authorise
OAuth into the ad accounts you want readable
3
Add server
Paste the MCP URL into your assistant
4
Ask
Start with an account snapshot

Step 1 — Sign up. Create an account at try.kiin.co. You will get the MCP server URL to paste into your assistant.

Step 2 — Authorise LinkedIn. You will be sent through LinkedIn's standard OAuth flow. Grant access only to the ad accounts you actually want readable. This is scoped: the server can see the accounts you authorise and nothing else. If you manage accounts for multiple clients, authorise them all now — cross-account questions are one of the main reasons to bother with this.

Step 3 — Add the server to your assistant.

In Claude: Settings → Connectors → Add custom connector, then paste the server URL. Claude will show you the tools it has picked up.

In ChatGPT: Settings → Connectors → Add, then paste the same URL. The flow differs slightly between plans and changes fairly often, so follow whatever the current screen says rather than a screenshot from six months ago.

You are not choosing tools here. The assistant decides which to call based on your question.

Step 4 — Ask something. Start with "give me an account snapshot for last month". If that returns real numbers matching what you see in Campaign Manager, you are connected.

💡 Verify before you trust it
Run one question you already know the answer to. Check last month's spend against Campaign Manager. If those two numbers match, the connection is live and scoped correctly. If they do not, you are looking at the wrong account or a stale token — and it is far better to find that out on question one than on question forty.

Your First Five Prompts

Most people connect it, ask "how are my campaigns doing", get a summary, and never come back. That is a waste. These five are where it starts earning its place.

1. The snapshot

"Give me an account snapshot for the last 30 days. Include spend, impressions, landing page clicks, leads and cost per lead."

Naming the metrics matters. Ask for "clicks" and you may get LinkedIn's all-inclusive click figure. Ask for landing page clicks and you get traffic.

2. The waste question

"How much of last month's spend went to job titles outside our ICP?"

This is the one that pays for the setup. On most accounts the answer is uncomfortable. Demographic data exists in Campaign Manager but is spread across screens and nobody reconciles it against the target list.

3. The format comparison

"Compare our thought leader ads against our single image ads on cost per landing page visit."

Comparing formats on the wrong metric is how thought leader ads get killed off for underperforming. On total clicks they look expensive; on landing page clicks they usually do not.

4. The warm accounts question

"Which companies engaged with our ads more than three times but never clicked through?"

These accounts never appear in a click report, because by definition they did not click. They are watching. This is the list your SDRs should be working, and it is the strongest argument for connecting ad data to an assistant at all.

5. The audit

"Run an audit and give me the three things to fix this week."

Ask for three. Ask for "an audit" and you get a wall of findings you will not action. Constraining the output is most of the skill.

Asking Questions That Get Real Answers

Here is the part the other setup guides skip.

LinkedIn's API returns a field called clicks. It counts likes, comments, shares, profile clicks and company page clicks alongside actual landing page visits. Most connectors pass it straight through as "clicks", and your assistant reports it as traffic.

On a thought leader ad the gap between the two is routinely 3–10×. Which means your CTR reads high, your CPC reads low, and both errors flatter the campaign.

⚠️ Fluent and wrong is worse than obviously wrong
A spreadsheet with a bad number looks like a spreadsheet. An assistant with a bad number writes you a confident, well-structured paragraph explaining why your campaign is performing well. Nobody re-checks the arithmetic on something that reads that cleanly. This is the single biggest risk of connecting AI to ad data, and it is a data problem, not a model problem.

Three habits that fix it:

Name the metric you want. "Landing page clicks", not "clicks". "Cost per landing page click", not "CPC". If the server distinguishes them, you get the right one. If it cannot, you have learned something useful about your server.

Ask which tool it used. "Which tool did you call and what date range did you apply?" A good session is auditable. If the assistant cannot tell you where a number came from, do not act on it.

Ask follow-ups. The first answer is a summary. The second and third are where the insight lives, because the assistant re-queries with the context of what you just learned rather than re-summarising the same snapshot.

The underlying principle is one we apply to every client report: clicks are not traffic, form opens are not leads, and video views are not attention. An assistant is only as honest as the field names it is handed.

When It Goes Wrong

The numbers do not match Campaign Manager. Almost always a date range or timezone difference, or the assistant is reading a different account. Ask it to state the account ID and date range it used.

Data has stopped updating. Expired LinkedIn token. This is the most common failure and it is silent — you keep getting answers, they are just old. Ask "is my LinkedIn connection healthy?"

It says it cannot find campaigns. Usually an authorisation scope issue: the account was not included when you granted access. Re-run the OAuth step.

It gives a vague answer. Usually the question was vague. "How are we doing" has no good answer. "What is our LP CPC by campaign for the last 30 days, worst first" does.

It cannot answer at all. Some questions are not answerable from ad data. "Why did our CPL go up?" may have nothing to do with LinkedIn — a landing page change will not show up in any ad report.

Where to Go Next

Once the connection is working, the return comes from asking better questions rather than more of them. The waste question and the warm-accounts question are the two that most often change what someone does that week.

For the full picture — what all 69 tools cover, how the eight LinkedIn Ads MCP servers compare, and what an automated audit actually checks — read the complete LinkedIn Ads MCP guide.

Frequently Asked Questions

How do I connect LinkedIn Ads to Claude?+
Sign up for an MCP server, authorise your LinkedIn ad accounts through OAuth, then add the server URL in Claude under Settings, Connectors, Add custom connector. Claude will list the tools it has picked up. Ask a question you already know the answer to, such as last month's spend, and check it against Campaign Manager to confirm the connection is scoped to the right account. The whole process takes about five minutes and requires no engineering help.
Can I connect LinkedIn Ads to ChatGPT too?+
Yes. Model Context Protocol is an open standard, not an Anthropic-only feature, so the same server URL works in ChatGPT. Add it under Settings, Connectors. The exact screens differ between plans and change fairly often, so follow the current interface rather than an older screenshot.
Do I need a developer to set this up?+
No. A remote MCP server needs no installation, no developer app and no terminal commands. If a setup guide asks you to run commands locally, that is a self-hosted server, which is a different and more involved approach. For LinkedIn Ads reporting there is no advantage to running one locally.
What LinkedIn permissions does it need?+
You need admin or campaign manager access on the ad account to complete the OAuth flow; viewer access is not sufficient to authorise. Access is scoped to the accounts you explicitly grant, so the server cannot see other accounts on your LinkedIn profile. If you manage several client accounts, authorise them together so you can ask questions across the whole portfolio.
Why do the numbers not match Campaign Manager?+
Usually one of three things: a different date range or timezone, the assistant reading a different ad account, or an expired LinkedIn token serving stale data. Ask the assistant which tool it called, which account ID and what date range it applied. If the connection itself is the problem, ask whether your LinkedIn connection is healthy. A silently expired token is the most common cause of data that quietly stops updating.
Why should I ask for landing page clicks instead of clicks?+
Because LinkedIn's clicks field counts likes, comments, shares, profile clicks and company page clicks as well as visits to your landing page. On thought leader ads the gap is routinely 3 to 10 times. If your assistant reports that field as traffic, your click-through rate is inflated and your cost per click is understated, and both errors make the campaign look better than it is. Naming landing page clicks explicitly avoids this, provided your server distinguishes the two fields at all.
Is it safe to give an AI assistant access to my ad account?+
Access is limited to the ad accounts you authorise. Read-only questions carry very little risk. If the server also exposes write tools that can create, update or pause campaigns, think about who on your team has access. Nothing runs autonomously in the background, and no tool raises a budget without an explicit instruction, but anything that would move real budget should be verified before you act on it.
What should I ask first?+
Start with an account snapshot for the last 30 days, naming the metrics you want including landing page clicks. Then ask how much spend went to job titles outside your ICP, and which companies engaged more than three times without ever clicking through. Those two questions surface things that are genuinely hard to see in Campaign Manager, and they are usually the ones that change what you do that week.

Connect your account in five minutes

Sign up, authorise your ad accounts, and start asking questions in plain English. Same tooling we use across 200+ managed B2B accounts.

Get the MCP Server