••
Insights & Resources
Expert strategies, case studies, and best practices for B2B marketing teams.
Ask Claude anything about your LinkedIn Ads, website visitors and CRM deals. DemandSense MCP
Private early access
Expert strategies, case studies, and best practices for B2B marketing teams.
Ask Claude anything about your LinkedIn Ads, website visitors and CRM deals. DemandSense MCP
Private early accessLinkedIn Ads
Revenue Attribution & Measurement
There’s a question most B2B marketers have been asking their dashboards for years without getting a straight answer: which ICP-fit companies saw our LinkedIn ads, came to the site, and still aren’t in our CRM?
It’s a reasonable thing to want to know. It’s also three systems away. LinkedIn Campaign Manager knows about impressions and clicks. Your visitor identification tool knows who showed up on the site. Your CRM knows who’s already a record. Nothing knows all three at once, so the question gets answered by hand, in a spreadsheet, badly, or not at all.
Model Context Protocol changes what you can do about that — but only if something joins the data before the model reads it. This guide covers what it takes to connect LinkedIn Ads to Claude, what DemandSense MCP gives Claude access to, the workflows it supports, and how to get into early access.
Here’s the short version before the detail.
Figures from a DemandSense MCP run against our own account. Account names are placeholders.
DemandSense MCP is in early access rather than general availability, so there’s no self-serve Claude Desktop config to paste in yet. The path in:
Worth knowing when comparing options: there’s no official LinkedIn MCP server, and most of the ones on the market give Claude direct access to the Marketing API and stop there. They’re useful for pulling campaign metrics. They aren’t the same category of thing as a server that resolves identity across three sources before answering.
The alternative most teams run today is an export, a spreadsheet and an afternoon.
| Manual CSV analysis | DemandSense MCP with Claude | |
|---|---|---|
| Getting the data | Export from Ads Manager, export from your visitor tool, export from the CRM | Already joined before the question is asked |
| Matching accounts | VLOOKUP on company name, then cleaning the near-misses by hand | Resolved at the person and company level in the product |
| Ad engagement depth | Clicks and conversions from the export | Impression-level exposure, including accounts that never clicked |
| How often it gets redone | Realistically monthly, because it’s an afternoon each time | Whenever you want to ask |
| Follow-up questions | New export, new spreadsheet | Next message in the same conversation |
Campaign Manager is built around the campaign, and B2B revenue happens around the account. That mismatch is the root of most reporting frustration on the channel.
Native LinkedIn reporting can tell you a campaign spent $8,000 and produced 40 leads. It can’t tell you that eleven of the accounts it reached most heavily are already open opportunities, that four are quietly reading your site every week, or that one is an existing customer you’re still paying to advertise to.
It’s also why wiring three separate connectors into Claude — LinkedIn here, CRM there, visitor tool over there — tends to disappoint. Each system holds the same company under a different key, and none of those keys matches another.
Figures from a DemandSense MCP run against our own account. Account names are placeholders.
With three separate connectors the model has to do that matching itself, across separate tool calls, with nothing shared between them. It will produce an answer and it will sound sure about it. The failure mode isn’t a refusal; it’s a plausible number you can’t check. Multi-touch attribution for B2B has the same problem one layer down.
Spend, impressions, clicks, CTR, CPC, CPM, conversions and lead gen form completions, by campaign, campaign group and individual creative. Every LinkedIn Ads MCP server reaches this layer. Our breakdown of LinkedIn ads metrics covers where each of these numbers misleads.
Which companies your budget reached, at impression level, and which came to the site afterwards — resolved to named people and companies with industry, employee size and region attached. That’s what turns “we reached 4,000 accounts” into “we reached these accounts, and these ones came looking.”
Visitor identification is probabilistic: match rates vary with your traffic and audience, and not every visitor resolves to a person.
With a CRM connected, deal stage, open pipeline value and account status land in the same record as the ad exposure — which is what makes LinkedIn pipeline attribution a one-turn question instead of a one-afternoon one. The server also works without a CRM connected; you keep the ad and visitor layers and lose the deal context.
| Data stream | What it contributes | Required? |
|---|---|---|
| LinkedIn ad engagement | Impression-level exposure per company, plus campaign and creative metrics | Yes |
| Identified website visitors | Named people and companies, firmographics, sessions, the campaign they arrived via | Yes |
| CRM state | Deal stage, open pipeline, account ownership, whether the account exists at all | Optional |
The familiar one, and the one an ad-only connector handles fine. Use Claude to compare this month against last, find campaigns whose CPC drifted, or explain where spend concentrated across all LinkedIn campaigns in the account. What the join adds is the second half of the question — whether the campaigns that look efficient are reaching accounts you actually want.
Filter the accounts your budget reached down to the ones matching your ICP, then look at what they did. The reverse cut is often more useful: which accounts are we paying to reach that have never once visited the site. That’s a spend problem hiding inside a delivery report.
Compare creatives on engagement rather than clicks, and look at frequency by account rather than by audience. In an ABM-shaped campaign against a finite list, what matters is how often one company has seen the same ad, not the blended average across a segment. Account-level LinkedIn ads benchmarks beat platform averages here, because a fatigue threshold that’s fine for a 500,000-person audience is not fine for 180 named accounts.
Brief yourself on a single account before a call: what they’ve been shown, how often, who from that company has been on the site, and what stage the deal is at. Or run it weekly — which ICP-fit accounts are showing ad and site activity this week, ranked by engagement. This is where buyer intent tracking stops being a category and starts being a list of names.
Ask which open opportunities saw your LinkedIn ads before they entered the pipeline, and how much open pipeline those accounts carry. The same joined record supports account-based marketing attribution without exporting anything, and it answers in accounts rather than last-click conversions.
Three questions cover most of it. Which accounts are we still paying to reach that already closed. Which have been heavily served and never engaged anywhere. Which CRM accounts went quiet but keep visiting the site after seeing our ads — ask Claude that last one and it usually turns up something a LinkedIn ads budget review would never catch, because the account looks dead in the CRM and alive everywhere else.
Every prompt below ran against a live account before it earned a place on the DemandSense MCP page. Treat them as a starting LinkedIn ads prompt library — edit the bracketed parts and analyze your LinkedIn Ads from there.
For agencies and consultants, that report prompt is the one that changes the week. The joined report becomes the deliverable, and it’s a list nobody else on the call can produce.
Most tools here are solving a real but narrower problem: getting LinkedIn ads analytics into a chat window faster. That’s an improvement over exporting a CSV, and several connectors do it well.
DemandSense MCP is aimed at the other thing — the questions that were never answerable at all, because the answer lived across LinkedIn, your website and your CRM at once. Joining those three streams into one buyer record before the model reads them is the whole product. The honest limits follow from it: read-only today, probabilistic visitor identification, early access in small batches with the team in the loop on every account.
If the three questions you ask your dashboard every Monday need two systems to answer, that’s what this is for — and it’s also why you’ve never had a straight answer to them. Early access is open at demandsense.com/mcp.
No. Once the MCP server is connected, you ask LinkedIn ads questions in plain English in a normal Claude conversation. There’s no query language, and no LinkedIn developer app to create. The underlying setup avoids a developer too — the DemandSense tracking code installs through Google Tag Manager.
No. DemandSense MCP is read-only today. Claude cannot edit campaigns, move budget, or launch or pause anything, and if you ask it to, it declines. Some MCP servers in the wider market do support write actions; this one deliberately doesn’t.
Within DemandSense MCP, LinkedIn is the ad channel. What it adds isn’t another ad platform but the two systems that decide whether LinkedIn spend worked. For comparing LinkedIn and Google Ads spend, or Google and Meta Ads side by side, use a general reporting connector.
Claude queries your live DemandSense workspace at the moment you ask, rather than a snapshot or training data. LinkedIn’s own reporting delay still applies to ad data exactly as it does in Campaign Manager, so treat same-day numbers as provisional.
The MCP server gives Claude access to your LinkedIn Ads data through your own DemandSense workspace and nothing else. Read-only means no path from a conversation to a change in your live LinkedIn account. It’s also built not to surface another client’s data or a named competitor’s private numbers, and not to compile personal details like home addresses or private phone numbers.
This matters most to agencies, and it’s one of the things early access is working through account by account rather than promising in advance. If you run several LinkedIn ad accounts across clients, say so on the application — that’s the shape of workspace the team wants in the current batches.
LinkedIn lead gen form completions sit in your campaign data alongside spend and clicks, so form performance by campaign and creative is a fair question. The more useful version is the joined one: of the accounts that filled in a form, which had been reading the site for weeks already — and of those that never filled one in, which were just as engaged.
Get expert insights and strategies delivered to your inbox weekly.
Book a personalized demo and discover how we can transform your B2B marketing.
See a quick demo video
AI Copilot
LinkedIn Ads
Learn what B2B teams should automate in LinkedIn Ads with AI, what decisions should stay human, and how AI-powered campaign management helps improve spend, targeting, and revenue impact.
AI Copilot
LinkedIn Ads
Learn how AI-driven marketing insights help B2B teams analyze LinkedIn Ads data, optimize campaigns, and make better budget, audience, and pipeline decisions.
AI Copilot
Learn how DemandSense AI Copilot helps B2B marketers analyze LinkedIn Ads data, diagnose performance changes, connect campaigns to pipeline, and get faster answers.