Analyze Campaigns with Claude: Your Daily Report via the LeadMetrics MCP Server

Every morning, the same routine: open the ad account, open the dashboard, open the CRM, compare numbers, write a summary for the team or client. With the LeadMetrics MCP server, Claude takes over that part. You ask in plain language, Claude pulls the numbers directly from LeadMetrics and writes the report - with leads, deals and revenue per ad, not just the numbers from the ad account.

This guide is for performance marketers, agencies and lead generation businesses. You'll learn how to connect Claude, what Claude can and can't see, and you'll get prompt templates for the most important analyses.

What you need

• A LeadMetrics account with at least one project that has ad platforms connected.
• Ideally a CRM or payment integration, so deals and revenue per ad are available. Without it, Claude can only analyze leads and costs.
• A Claude account that lets you add custom connectors.

Connect Claude to LeadMetrics

The Model Context Protocol (MCP) is an open standard that lets AI assistants access external tools. LeadMetrics provides an MCP server for this, which you add to Claude as a connector.

1. Add the connector

   Open the connector settings in Claude and add a custom connector with the URL https://mcp.lead-metrics.com/mcp.

2. Sign in

   Sign in via OAuth. Alternatively, create an MCP API key in LeadMetrics under your organization settings.

3. Test the connection

   Ask Claude: "List my LeadMetrics projects." If your projects show up, you're connected.

What Claude can see via MCP

The MCP server gives Claude a few clearly scoped tools:

For reports, getPerformance is the key tool. With it, Claude can:

• query up to 20 metrics at once, for example spend, leads, deals, cost per acquisition, revenue and ROAS,
• filter to one platform or drill from a campaign into its ad sets and ads,
• choose between first-click attribution and last-click attribution,
• limit results to currently active ads.

The limits matter too:

• No personal lead data. The MCP server doesn't pass on names, email addresses or phone numbers of your leads.
• No images or videos. Claude sees the names of campaigns, ad sets and ads, but not the creatives themselves. That's why an ad naming convention pays off.
• No changes in the ad account. You still change budgets and ad status yourself.

The morning report: a prompt template

You can copy and adapt this template directly. It works best when you include your targets.

Create a campaign report for the LeadMetrics project "Mortgage".

Period: the last 7 days, compared with the 7 days before.
Attribution: last click.
Metrics: spend, leads, cost per lead, deals,
cost per acquisition, revenue, ROAS.

Our targets: cost per acquisition below $600, ROAS above 3.

Structure:
1. Three-sentence summary: what got better, what got worse?
2. Table per campaign with change vs. previous week.
3. The 3 best and 3 weakest ads by cost per acquisition
   (only ads with at least $300 spend).
4. Anomalies: ads with spend but no leads.
5. At most three concrete suggestions. Flag where the data
   is still too thin for a decision.

Claude first calls listProjects and listMetrics to find the project and metrics, then queries getPerformance twice - once per period. Claude calculates the comparison itself.

Save instructions permanently

So you don't have to retype targets and structure every time, create a project in Claude with fixed instructions. Include your KPIs and targets, your break-even ROAS, the attribution model, minimum spend for an evaluation and the report structure you want. After that, a short prompt like "report for yesterday" is enough.

More prompt recipes

Winners and losers at ad level

Show me all active Meta ads in project "Nurse Recruiting" for the
last 30 days, sorted by cost per acquisition.
Which ads bring cheap leads but no deals?

This finds ads that look good on cost per lead but bring nothing in sales.

Drill down from campaign to ad

Which campaign had the highest ROAS in September?
Then go into its ad sets and ads and explain
where the difference comes from.

Compare first click and last click

For project "Coaching", compare deals per campaign over the last
60 days with first click and with last click. Which campaigns
bring new contacts, and which tend to close?

Campaigns that look much better with first click bring people in for the first time. Switching them off because they look weak in the last-click report often costs deals later.

Client update for agencies

Write a short client update for each of my projects:
spend, leads, deals and ROAS last month, change vs. the previous
month, one sentence of context. Factual, no jargon.

For the formal report with cover page and pages per campaign, keep using the PDF reports and report automation in LeadMetrics. The Claude update complements them as a short message.

Include custom metrics

Which custom metrics exist in project "Real Estate"? Analyze
booked meetings per campaign and calculate cost per meeting.

Reading the results correctly

Claude calculates reliably, but the interpretation depends on the same pitfalls as any reporting:

• Small numbers: Two deals versus one isn't a trend. Set a minimum amount of spend or leads in your instructions before Claude evaluates.
• Sales cycle: Leads from the last few days don't have deals yet. Evaluate cost per acquisition for periods that lie at least one typical sales cycle back.
• Differences from the ad account: LeadMetrics counts leads and deals from your tracking and CRM, not by the platform's attribution window. Differences from Meta or Google are normal. The guide on attribution windows explains why.
• Check suggestions: Claude only knows your margin, sales capacity and strategy if you tell it. Budget decisions are yours.

Copilot in LeadMetrics or Claude?

Both access the same data but suit different jobs.

If you want to grow Claude into the analysis agent of a larger setup, the guide on the AI marketing team shows which roles make sense and where humans should stay involved.

Data protection

The MCP server doesn't pass personal data of your leads to Claude, nor customer names from projects. Write actions such as creating a project are logged. Claude gets exactly what it needs for campaign analysis: metrics per channel, campaign, ad set and ad. More on the feature set on the AI Marketing Copilot page.