Claude Unpacked: MCPs

This is the third and final part of the Claude Unpacked series I’ve been doing to help you get more out of using AI in marketing.

Previous parts were slash commands and Skills, then subagents. This week, it’s MCPs.

MCPs sound technical, but I’d say it’s no more difficult than setting up a website domain - especially when AI can do the heavy lifting for you.

So in this final part I’m looking at what an MCP is and 4 that you can get started with.

Side note - if you have found this series useful, do me a favor and share the link.

MCP stands for Model Context Protocol. It’s an open standard that lets Claude connect to external tools and data.

Normally, Claude can only work with the information available in the conversation, the files you upload, and its built-in capabilities. An MCP extends that access

Depending on the platform, it might let Claude retrieve live data, search a database, run an audit, create an asset, or make changes inside another tool. Each MCP allows a specific set of actions, so Claude can only do what that connection allows.

You may also see the terms connector and plugin:

I recommend keeping permission checks enabled for sensitive actions such as publishing, deleting, or updating files.

Adding one takes about two minutes: Settings > Connectors > Add custom connector > paste the URL > authorize.

Ahrefs MCP gives Claude direct access to Ahrefs data, including keywords, rankings, competing pages, backlinks and Site Audit projects. It’s on Lite plans and up, and every request draws from your monthly API units.

Instead of finding and comparing all the data yourself, Claude can move between the datasets in minutes.

For example, if you were creating a search-led article or landing page. You could give Claude your product, audience, and campaign goal, then ask it to use Ahrefs to:

You skip the export-and-paste step entirely; the research happens inside the content workflow.

It gets more useful as a recurring Claude Routine (Claude Code’s scheduled runs, capped per day depending on your plan).

For example, every Monday it could check your most important pages against the previous period and flag:

Then the output can be an email to yourself with the report, if you have Gmail connected.

Of course, this is a basic use case; you could go further with a fully automated run using other MCPs.

I have an automated run that uses Search Console MCP to identify changes > WordPress MCP to go and access the content and audit > Ahrefs MCP to understand the keywords/competition > Claude then adjusts the article draft… and so on.

Higgsfield MCP gives Claude direct access to Higgsfield’s image and video generation models.

Claude can choose the model, configure the generation, and work with the finished assets inside the same workflow, so there’s no writing a prompt in Claude, copying it into another platform, and returning with the results.

The strongest use case is producing and testing paid social creative.

Connect Higgsfield alongside Meta Ads MCP, then give Claude your product information, brand rules and recent campaign results.

You could ask it to:

Use Higgsfield to generate batches of UGC-style videos, product demonstrations, static ads and short cinematic spots.

Upload the approved assets into paused Meta campaigns for review.

The value is in the loop it creates.

For example, a weekly Claude Routine could review the previous 7, 14, and 30 days of performance, identify which creatives are declining, generate variations of the strongest concepts, and queue the next batch for approval.

Other useful workflows include:

Royal MCP gives Claude structured read-and-write access to WordPress.
It can work with posts, pages, media, taxonomies, metadata and revisions, with additional tools available for systems such as Elementor, ACF and WooCommerce.

Royal MCP also includes authentication, rate limiting, and activity logs because the connection can make changes to the website.

Where it earns its keep is handling the entire content-update process.

For example:

You could also run a scheduled task that reviews pages matching defined conditions, such as outdated dates, broken links, or incomplete SEO fields, and prepares the corrections as drafts.

It’s also good for:

Because the theme files already exist, it doesn’t need much guidance to stay aligned with your site/brand. I had Claude build 20 landing pages from a basic prompt, and it was done in 15 minutes - 90% ready to publish. Saving me hours.

I’m also testing a fully automated content publishing workflow with this MCP. All I do is read the published content to check for major issues. That’s coming in a future newsletter.

Screaming Frog’s MCP lets Claude control the SEO Spider, analyze crawl results and run Node.js scripts against the data. It’s the one here that isn’t click-and-connect: you need a paid license, v24 or later, and the app open on the same machine.

That means Claude can join crawl datasets, apply custom conditions, compare crawls, and turn the result into a report based on your needs.

You could schedule Screaming Frog to crawl your site weekly, then have Claude compare the new crawl against the previous one and flag changes:

If Search Console, analytics, or URL Inspection data is configured inside Screaming Frog, Claude can combine those with the crawl.

For example, it could find articles with declining clicks, bad internal linking, and declining index coverage, then separate likely technical problems from pages that might need a content update.

Also worth trying:

That’s the final part of Claude Unpacked.

Skills give Claude reusable instructions.
Subagents give it specialist workers.
MCPs give it access to the external tools and data.

Start with one tool you already jump between in your workflow and one part of the process where you constantly move information between platforms.

Once you get the hang of connecting one MCP into a workflow, try adding more and automating everything.