Skip to content
  • There are no suggestions because the search field is empty.

Using the MCP integration to setup and run your marketing with an AI

Once you have connected Pearl Diver to Claude or any MCP-compatible AI tool, how to build, manage, and explore your intent audiences using natural language and then manage your ad campaigns.

Getting Started with Pearl Diver and Claude: From Sign-Up to Your First Campaign.

This is a beginner's guide for marketers who are new to audience data, AI connectors, and paid advertising. No prior experience assumed.

Plan on about 30 to 45 minutes to go from zero to a live-ready audience.

What you'll have by the end

By the time you finish this guide, you'll be able to:

  • Log into Pearl Diver and Claude and connect the two together

  • Ask Claude, in plain English, to build a target audience from Pearl Diver's data

  • Push that audience into Meta or Google Ads

  • Launch a first campaign set up to give you useful results quickly You do not need to write any code, and you don't need a marketing background. 

The workflow at a glance

Here is the overview of the steps. The rest of the guide walks through each step in detail.

  1. Create your Pearl Diver account and get your access credentials

  2. Sign up for (or log into) Claude

  3. Connect Pearl Diver to Claude as a custom connector

  4. Build your first audience by describing it to Claude

  5. Send the audience to Meta or Google Ads

  6. Launch a starter campaign and read the results

Three key ideas to understand first:

Skim these before you start. They make every step afterward easier.

Intent and segment data.

Most advertising fails because it shows the right ad to the wrong people. Segment data groups people by shared characteristics (life stage, income band, homeownership, and so on). Intent data goes a step further and reflects what people appear to be actively interested in or in-market for right now. Pearl Diver's job is to turn these signals into a defined audience, a list of the kinds of people you actually want to reach.

What an MCP connector is.

MCP (Model Context Protocol) is an open standard that lets an AI assistant like Claude plug into an outside service and use it directly. When you connect Pearl Diver to Claude, Claude can query Pearl Diver's data for you during a normal conversation. No dashboards, no exports, no copy-paste. You ask; Claude fetches and builds.

Why Claude plus Pearl Diver.

On its own, Claude is a smart assistant with no access to your audience data. Pearl Diver on its own is powerful but expects you to know what to ask it. Put them together and you can describe who you're trying to reach in everyday language, for example "homeowners in Texas who recently refinanced", and Claude translates that into a real, sized, exportable audience. Plus this process works with the free version of Claude.

Step 1: Create your Pearl Diver account

Head to pearldiver.blackpearl.com and sign up (or log in, if you already have an account).
That's really all you need here. There's no API key to copy or manage.

The login you create at this step is the same login you'll use to connect Claude in Step 3, so keep those credentials handy.

Step 2: Get into Claude

If you don't already have an account, go to claude.ai and sign up. A free account works to try
this out, but note the limits below.

A quick note on plans, because it affects the connector step:

  • Free accounts can add one custom connector. Enough to try Pearl Diver, but nothing else alongside it.

  • Pro or Max (individual paid plans) let you add multiple custom connectors and are the smoothest experience for regular use.

  • Team or Enterprise accounts can use custom connectors too, but an account Owner has to add the connector for the organization first (covered in Step 3).

If you're just evaluating Pearl Diver, Free or Pro is the fastest way to start.

Why we suggest Claude first: Claude supports connectors on every plan, including the free one, with full audience-building included. You can connect Pearl Diver, build and save audiences, and run the whole workflow in this guide without paying for anything. That is the main reason we point people to it.

Already paying for anotherAI assistant?

Pearl Diver's connector is built on the open MCP standard, so it isn't tied to Claude alone. Other assistants that support custom connectors can use it too, and it runs the full build-and-save workflow on enterprise-grade plans (we've confirmed this on ChatGPT Enterprise, for example). One thing to be aware of: on individual paid plans, custom connectors can be limited to reading data rather than creating and saving audiences, so a business or enterprise workspace gives the complete experience. The steps in this guide are written for Claude, which is the smoothest path from a standing start and works on every plan, including Free.

Step 3: Connect Pearl Diver to Claude

This is the one technical-feeling step, and it takes about two minutes. You're telling Claude,  "here's an outside service I want you to be able to use."

The Pearl Diver connection address (MCP server URL):
https://mcp.pearldiver.io/mcp

If you're on a Free, Pro, or Max plan

  1. In Claude, open Customize > Connectors (also reachable through Settings > Connectors).

  2. Click the add button next to Connectors and choose Add custom connector.

  3. Give it a name, for example Pearl Diver, and paste in the URL https://mcp.pearldiver.io/mcp

  4. Click Add, then click Connect.

  5. You'll be taken to a Pearl Diver sign-in page. Log in with the same credentials you  created in Step 1, and approve the connection. That's it, there's no token to paste.

(The MCP URL also lives inside Pearl Diver under Settings > Integrations, but you don't
need to go there. The address above is all you need.)

If you're on a Team or Enterprise plan

An account Owner adds the connector once for the whole organization:

  1. Go to Organization Settings > Connectors > Add > Custom.

  2. Enter the name and the URL above, and save.

  3. Each team member then opens Connectors, finds Pearl Diver (it'll be labeled Custom), clicks Connect, and signs in with their own Pearl Diver credentials.

This keeps each person's access limited to the data their own account is allowed to see.

Turn it on and check that it worked

Custom connectors start with permissions set to "Ask each time", so Claude will ask before
it uses a Pearl Diver tool the first time. That's expected. You can enable a connector for a
specific conversation using the + button in the lower-left of the chat box.

To confirm everything's wired up, start a new conversation and ask Claude something
simple like:

"Using Pearl Diver, what audiences can you see, and what kinds of data can I filter people by?"

If Claude comes back with a list of your existing audiences or the available data fields, you're connected. If it says it can't find the tools, see Troubleshooting at the end.

For Other AI platforms

You will need to have one of their paid plans that allows creating custom connectors, for any enterprise ones your IT team will typically set up the connector and you will just click "add" once it's available.

Step 4: Build your first audience

Now the fun part. You describe who you want to reach, and Claude uses Pearl Diver to build
and size the audience. There are two natural starting points.

Good to know: Claude can build and save two kinds of audiences through the connector.
Intent audiences, built from intent topics, industries, and keywords, and segment
audiences, built from characteristics like income, location, or homeownership. Webvisitor audiences are the exception. Those are still built and edited directly in the Pearl
Diver dashboard.

A note on sensitive data: some categories, such as health, finances, or anything tied to a protected characteristic, need extra care, and ad platforms often restrict how you can
target or what claims you can make around them. Claude will flag these as they come
up, but it's worth keeping in mind as you describe who you want to reach.

Path A: Start from who you want to reach

Best when you already have a picture of your ideal customer. Just describe them. Be specific about the things that matter and let Claude ask about the rest.

Example prompts:

  • "Build me an audience of homeowners in Florida, aged 35 to 55, with a household
    income above $100k, who look like they may be in the market to refinance."
  • "I sell premium home-security systems. Help me build an audience of likely buyers and tell me roughly how big it is."

Claude will translate your description into Pearl Diver's data fields, build the audience, and report the estimated size. If a filter you asked for isn't available, it'll tell you and suggest the closest match.

Path B: Start from what's already working

Best when you have an ad, offer, or existing customer profile that's performing and you want more people like that. You're essentially reverse-engineering the audience from a proven result.

Here's the move most people don't think to try: you can hand Claude the ad itself. Because Claude can read an uploaded image or pasted text, you don't have to describe your audience from scratch. You can show Claude what's already working and let it help you figure out who to target.

Where Claude really earns its keep here is the thinking part: intent audiences are built by choosing the right topics, industries, and keywords, and knowing which to pick is the hard bit for a beginner. Claude reads your creative, works out who it's speaking to, and suggests the intent topics that match. Then you turn those into an audience.

Example prompts:

  • "[Upload your ad image, or paste the ad copy] This is my best-performing ad. Read it, work out who it's really speaking to, and suggest the Pearl Diver intent topics, industries, and keywords I should target to reach more people like that."

  • "My best-performing ad is a discount offer for first-time homebuyers. What intent topics and keywords should I be selecting to build an audience that resembles that group?"

  • "Here's a description of my current best customers: [paste description]. Which topics would you start with, and why?"

Once you and Claude have landed on a shortlist, Claude builds and sizes the audience for you directly through the connector. There's no need to open the dashboard and recreate it by hand.

Size it, then refine

Two things to check before you move on:

Is it big enough to advertise to? Very small audiences can be hard for ad platforms to deliver against, and some platforms won't run against a list below a minimum size. If Claude reports a small number, ask it to broaden. Loosen an income band, add a neighboring state, or relax a filter.

Is it too broad?

If the audience is enormous and generic, you'll waste budget. Ask Claude to tighten it toward your best-fit customer.

Good prompt for this back-and-forth:

"That audience is smaller than I'd like. What could I loosen to grow it without losing the people most likely to convert?"

Save it, no dashboard required

When you're happy with the definition and size, just ask Claude to save it:

"That's the one. Save it to Pearl Diver as [name]."

Claude creates and saves the audience straight into your Pearl Diver account through the connector. It names it, stores the final size, and asks which of your connected platforms (Meta, Google, and so on) you'd like to attach it to, so it's ready to activate. The saved audience shows up in your Pearl Diver dashboard right away, exactly as if you'd built it there by hand. You can open it, review it, or keep working with it in Claude.

Give it a Meta-friendly name. Meta now flags custom audience names that hint at sensitive topics like health or income, and a flagged audience stops working. It can't be used in campaigns, can't grow, and can't seed a lookalike. This can catch perfectly legitimate audiences out purely because of a word in the label. The name is just a label and doesn't change who is in the audience, so keep it neutral: name it for the business segment, abbreviate anything sensitive, and leave targeting details like age bands out of the name entirely (the audience's own filters handle those).

For example:

  • "Health and Wellness, aged 35 to 55" becomes "HW Prospects"

  • "High Income Homeowners" becomes "Tier 1 Homeowners"

  • "Supplements and Vitamins Q3" becomes "SV Q3"

Neutral names like "Prospecting list Q3 2026" or "Product page visitors" are the goal. If you're unsure, just ask:"Give this audience a neutral, Meta-safe name."

Want to change it later? Just ask:

"Open my [name] audience and widen the income band a little."

Claude reopens the saved audience and updates it for you. No manual editing.

(Connecting an audience to an ad platform assumes that platform's integration is already set up. Step 5 covers that one-time step.)

Prompts you might not think of

The examples above are the obvious ones. These are the moves that separate people who chat with Claude from people who work with it. Copy, paste, and adapt.

  • Ask what's possible before you start. "Before I build anything, what can I actually filter people by in Pearl Diver?" Saves you from chasing filters that don't exist.

  • Point Claude at a web page. "Here's my product or landing page: [URL]. Based on what you read there, build the audience most likely to buy."

  • Get three tightnesses at once. "Give me a broad, a medium, and a tight version of this audience, each with its estimated size, so I can pick the trade-off."

  • Have Claude pressure-test your audience. "Look at this audience. Who am I probably excluding that I shouldn't, and who's in here that likely won't convert?"

  • Close the loop the other way. "Now that the audience is built, write me three ad headlines that would land with these specific people." One conversation, both the audience and the creative.

  • Don't forget exclusions. "Build this audience but exclude people who resemble my existing customers." Almost everyone forgets this and pays to re-reach people they already have.

  • Set up an A/B test in one go. "Give me two competing audience hypotheses for this product so I can test which one converts better."

A useful habit: when Claude proposes an audience, ask "why these filters?" It'll explain its reasoning, which is often where you spot the assumption you'd want to change.

Step 5: Send the audience to your ad platform

An audience only earns its keep once it's inside the platform where you run ads. Pearl Diver connects to the major ad platforms so you can push your audience straight across. Once a platform's integration is set up (the one-time steps below), Claude can attach any audience
it builds to that platform for you, so activation becomes part of the same conversation.

Follow the guide for the platform you use:

You'll want the ad account on the platform side connected first. Each guide above walks through that. Once the audience is pushed across, it lands inside the platform as a matched audience you can target directly, and, more powerfully, as a seed the platform can use to find more people like them (that's the next step).

Step 6: Launch paid ads that get results early

You don't need to be an expert to get a solid first result. You need a good audience (done), a clear offer, and a simple, disciplined setup. Here's the beginner's playbook.

The one idea that matters most

Your Pearl Diver audience is valuable in two ways. You can target it directly, but it's often more powerful as a seed for the platform's "find more people like these" feature. On Meta that's a Lookalike Audience; on Google it's the "similar" or optimized targeting the platform builds from your Customer Match list. A tightly-defined seed produces a better lookalike, which is exactly why starting with quality Pearl Diver data gives you an edge over guessing.

Meta (Facebook and Instagram)

  1. Load your audience. Your activated Pearl Diver audience appears in Meta's Audiences section as a Custom Audience.

  2. Create a Lookalike from it. Start with a 1 to 3% lookalike for your country. Smaller percentages mean closer resemblance to your seed.

  3. Put your creatives together and let Meta optimize. Keep it to one campaign with a single ad set, and load 3 to 5 different creatives into it. You don't need to split them out or manage them individually. Meta's delivery system automatically shifts budget toward the creatives that perform and away from the ones that don't. Over-splitting into many ad sets just starves each of data.

  4. Pick the right goal. Choose the objective that matches what you actually want (leads, purchases, traffic). The platform optimizes toward whatever you tell it.

  5. Set a budget you can run for at least a week. Meta needs roughly 50 conversions per week per ad set to optimize well. If your budget can't reach that, keep everything in the one ad set rather than spreading it thin.

One important choice about lead forms. If you run a Meta lead form (an instant form that collects details inside Facebook), those people never visit your website, so Pearl Diver can't identify, enrich, or retarget them afterwards. If you want to get the full value of Pearl Diver, including its ability to recognise and re-engage your site visitors, send traffic to your own website rather than to an on-Facebook form.

A simple two-week starter plan (Meta)

Most Pearl Diver users run on Meta, so this plan assumes it. The golden rule on Meta is to let the algorithm do the sorting. It's genuinely good at it, and every manual edit risks resetting the "learning phase" and setting you back.

  • Days 1 to 3: Launch with your lookalike audience and all 3 to 5 creatives in one ad set. Then leave it completely alone.

  • Days 4 to 7: Let it run. Early numbers are noisy, and Meta is still learning. Resist every urge to tweak, you'll only reset its progress.

  • Days 8 to 10: Now look at results. Meta will already be favoring the stronger creatives on its own. You don't need to pause the weak ones; the system is handling that.

     

  • Days 11 to 14: Once you can see a clear winner or two, the useful move isn't pausing losers, it's feeding the machine more of what works. Add a couple of fresh creatives in the same vein, or test a second audience variation.

The hardest discipline for beginners is waiting. On Meta especially, doing less is usually the
right call.

A note on other platforms: this hands-off, single-ad-set approach is built around Meta's automatic optimization. On some other platforms you may play a more active role in managing what runs, so check the specifics for whichever platform you're using.

If you're running on Google (or another platform)

The same core idea carries over, use your Pearl Diver audience as a seed and let the platform expand it, but a few specifics differ:

  1. Connect your audience as a Customer Match list (Google) or the platform's equivalent.

  2. Let the platform expand it. Turn on optimized or similar targeting so it finds users resembling your list rather than only exact matches.

  3. Choose a campaign type by goal. For beginners on Google, a Search campaign (for people already searching) or a Performance Max or Demand Gen campaign (for broad, automated reach) is the simplest start.

  4. Feed the system good inputs. This automation is only as good as your creative and your conversion tracking, so make sure conversions are tracked before you spend meaningfully.

Unlike Meta, some platforms expect a more hands-on role in managing and pausing what runs, so lean on that platform's own guidance for day-to-day management.

Metrics that actually matter at the start
Don't drown in dashboards. Watch these:

  • Cost per result (cost per lead or per purchase), the number that tells you if it's working.

  • Click-through rate (CTR), low CTR usually means the creative or audience is off.

  • Conversion rate, if clicks are cheap but nobody converts, the problem is your landing page or offer, not the ads.

Close the loop: let Claude read your results

Everything so far has been about building the audience and launching. The bigger advantage comes once the campaign is running, when Claude can see how it's actually performing.

Just as you connected Pearl Diver in Step 3, you can connect Claude to your ad platform, for example Meta or Google, using that platform's connector, added the same way under Customize > Connectors. Once it's linked, Claude can pull your live campaign numbers and give you specific, data-based advice instead of general tips.

With your results in front of it, Claude can:

  • Spot which audiences, creatives, or campaigns are winning and which are wasting budget

  • Compare cost per result across campaigns so you're judging like for like

  • Flag when a campaign is still in the learning phase and simply needs more time

  • Suggest concrete next moves, such as which creatives to refresh or which audience to scale

Example prompts:

  • "Using my Meta account, pull the results for my Pearl Diver campaign and tell me what's working and what isn't."

  • "Compare cost per lead across my active campaigns and explain why they differ."

  • "This campaign's cost per lead looks high. Based on the numbers, what would you change first?"

A quick tip: results only tell the full story when you look past cost per lead to what those leads are actually worth. If you can, tell Claude your conversion or booking rate so it can weigh cost against value, not just volume.

(If your ad platform isn't available as a connector yet, you can still paste in a screenshot or an export of your results and ask Claude to assess them. It reads those just as well.)

Common beginner mistakes to avoid

  • Judging too soon. Give campaigns a week before drawing conclusions.

  • Fiddling during the learning phase. On Meta, edits reset the algorithm's learning. Set it up well, then leave it alone. When you do test something new later, change one thing at a time.

  • Too many tiny audiences. Fragmenting budget across many small audiences starves each of the data it needs.

  • Weak creative on a great audience. A perfect audience can't rescue a boring ad. Put real effort into the message and image.

  • No conversion tracking. If you can't measure results, you're flying blind. Set this up first.

Troubleshooting and FAQ

Claude doesn't seem to see Pearl Diver's tools. Make sure the connector is enabled for the conversation (the + button in the chat box), that you completed the Connect or authenticate step, and that your credential is still valid. Disconnecting and reconnecting the connector resolves most cases.

The connection fails when I click Connect. Double-check the URL is exactly
https://mcp.pearldiver.io/mcp with no extra characters, and that your access credential is
entered correctly underAdvanced settings.

My audience is too small to run ads. Ask Claude to broaden the definition. Loosen an income or age band, widen the geography, or relax the most restrictive filter. Then re-check the size.

I'm on a Team or Enterprise plan and can't add the connector. Only an account Owner can add a custom connector for the organization. Ask your Owner to add it, then connect with your own login.

Can I reuse an audience I built earlier? Yes. Ask Claude to list your saved audiences, or to open a specific one by name, and continue from there.

Glossary

  • Audience: a defined group of people you want to reach, built from data filters.

  • Custom Audience (Meta) or Customer Match (Google): an audience you supply to the ad platform to target directly.

  • Lookalike or Similar audience: a new, larger audience the ad platform builds by finding people who resemble your supplied audience.

  • Connector (MCP): the link that lets Claude use an outside service like Pearl Diver.

  • Creative: the actual ad, the image or video plus the words.

  • Intent data: signals about what people appear to be actively interested in or shopping for.

  • Segment data: groupings of people by shared traits (income, life stage, location, and so on).

  • Seed audience: the starting audience a platform uses to build a lookalike.