How We Run an End-to-End Content Cluster with AI Agents (Our Own MCP Workflow)

Agentic SEO turns content cluster planning from a two-week manual slog into a workflow an AI agent runs from a single prompt. Instead of you researching keywords, drafting a brief, writing an article, and then remembering to link it back to related posts, an agent connected through MCP handles the mapping, drafting, and linking in one continuous pass.

Most founders running their own marketing don't lack ideas. They lack the hours to turn one idea into a dozen connected articles that actually build topical authority.

Here's the workflow: connect an agent, give it a topic, and let it map, draft, link, and monitor a cluster while you focus on the parts that actually need a founder's judgment.

By the end, you'll know exactly what to hand off to the agent and what to keep for yourself.

• Topical authority comes from interlinked clusters, not isolated posts, so plan the cluster before writing a single article.
• An MCP server connects your AI agent (Claude, Cursor, or similar) directly to content-planning tools without manual copy-pasting.
• Setting a voice profile before generation prevents the generic "AI blog" tone that readers and search engines both discount.
• The agent should map keywords and propose links, but a founder should still pick the pillar topic and approve the final linking structure.
• Internal linking isn't a finishing touch. It's the mechanism that tells search engines your cluster is one coherent topic, not a pile of unrelated posts.
• Credit-based generation lets you scale output without hiring, but it doesn't replace judgment on brand voice or strategic priorities.
• A cluster with weak internal links can get crawled and still never get indexed, no matter how good the individual articles are.

Publish a blog post. Share it once. Watch it flatline in Search Console for months. This is the default experience for most solo founders and small teams handling their own content, and it's not because the writing is bad.

Topical authority is the search engine's confidence that your site is a genuine, comprehensive resource on a subject, built through a network of related, interlinked pages rather than a single article. A lone post about "email onboarding sequences" tells Google almost nothing about whether your site understands email marketing broadly. A pillar page on email marketing, linked to supporting articles on subject lines, send timing, and deliverability, tells a completely different story.

So why does random posting fail? Not just stylistically, structurally. Search engines evaluate pages partly through the links pointing to them, both external and internal. A standalone post has no internal link equity flowing in. It sits on an island. Even a well-written, well-researched article can take a long time to rank, and much of what currently ranks well has been live for years, not months, according to Ahrefs' analysis of 2 million keywords, which found only 1.74% of new pages reach the top 10 within a year and that 72.9% of current top-10 results are three or more years old (Ahrefs, How Long Does It Take to Rank, 2025).

That data point matters more than it looks. If new content rarely cracks the top 10 within a year regardless, the sites that eventually win are the ones building depth and internal connections while they wait, not the ones hoping one post gets lucky.

So the fix isn't writing harder. It's writing connected.

Connecting Claude or Cursor to the Inbounder MCP Server

An MCP server is a connection layer that lets an AI agent like Claude or Cursor talk directly to an external tool's functions, rather than you manually exporting data and pasting it back and forth. Once connected, the agent can call functions like "map keyword cluster" or "generate brief" the same way it calls a file-read function in your codebase.

Setup is a configuration step, not a coding project:

1. Add the Inbounder MCP server endpoint to your Claude Desktop or Cursor MCP config file.
2. Authenticate with your Inbounder account credentials.
3. Confirm the connection by asking the agent to list available tools. It should return functions for keyword mapping, brief generation, drafting, and link auditing.

If you're comparing this setup against other options in the market, SEO MCP servers compared across GSC, Ahrefs, Semrush, and Inbounder breaks down what each one actually does versus what it's marketed to do.

Setting Your Voice Profile So Drafts Stay On-Brand

A voice profile is a stored set of instructions covering tone, vocabulary, sentence structure preferences, and banned phrases that the agent references on every generation request. Skip this step and you'll get technically correct, personality-free drafts that read like every other AI-generated post published this year (there are a lot of them now, and they all sound the same).

Feed the agent three or four pieces of your existing content, along with explicit notes: "avoid corporate jargon," "keep sentences short," "we say 'customers,' never 'end users.'" This profile gets referenced automatically on every future draft, so you're not re-explaining your brand voice for article number nine.

How the Agent Surfaces Related Keywords and Groups Them by Intent

Type a seed topic. Something like "customer onboarding for B2B SaaS." The agent queries keyword data, pulls back a wide set of related terms, and groups them by search intent (informational, comparison, or transactional) rather than dumping an unsorted list on you.

This matters because intent grouping is what turns a keyword list into an actual content plan. A term like "customer onboarding checklist" signals someone wants a practical resource. A term like "customer onboarding software" signals someone comparing tools. Mix those into a single article and you confuse both the reader and the search engine trying to figure out what the page is actually for. If you want the mechanics of this step in more depth, the AI agent keyword research workflow from topic to prioritized list walks through exactly how the prioritization logic works.

Picking the Pillar and Supporting Topics from Agent Output

The agent proposes a structure: one pillar page covering the topic broadly, plus supporting articles addressing specific sub-questions. This is where founder judgment matters most in the entire workflow. The agent doesn't know your product roadmap, your sales team's most common objection, or which sub-topic your competitors have already saturated. You do.

Review the proposed cluster and reorder priorities based on business context, not just keyword volume.

What the Agent Hands Off vs. What You Review

Once the cluster structure is approved, the agent generates a brief for each article, covering target keyword, search intent, recommended headings, and competing content it should differentiate from. From there, it drafts the full article using your voice profile.

What you should review before publishing:

• Factual claims and statistics, especially anything specific to your industry or product.
• Examples and case studies, to confirm they're relevant and not generic filler.
• The opening 100 words, since this is what both readers and AI answer engines scan first.

What you generally don't need to re-check line by line: sentence structure, heading hierarchy, and basic formatting, since the agent handles these consistently once the voice profile is set.

How Links Get Proposed and Where You Approve Them

This is the step most manual workflows skip entirely, and it's also the one doing the heaviest structural lifting. The agent scans every article in the cluster, identifies contextually relevant anchor points, and proposes links between pillar and supporting content, plus links between supporting articles that share overlapping subtopics.

You approve or reject each proposed link rather than the agent publishing changes blind. This matters because a bad internal link (one stuffed awkwardly or pointing to a barely-related page) does more harm than no link at all. For a deeper look at how automated link audits catch gaps that manual reviews miss, see automated internal-linking audits and how agents find and fix link gaps.

Skip this step and you risk the exact failure mode that sinks otherwise solid clusters: pages that get crawled but never indexed, because nothing in the site's structure signals they belong to a coherent topic. Sidebar links alone don't do this job. In-body, contextual links are what search engines weight as genuine topical signals.

How Credit Usage Maps to Articles, Briefs, and Audits

Instead of paying for a writer's hours or a full-time SEO hire, credit-based generation ties cost directly to output. A brief consumes one allotment of credits. A full draft consumes more. A link audit across an existing cluster consumes its own separate amount.

This model works because it scales down as easily as it scales up. A solo founder testing one cluster spends a small amount to validate the topic before committing further. Meanwhile, an agency running clusters for multiple clients scales usage proportionally without renegotiating a retainer every time volume shifts.

Monitor performance the same way you'd monitor any content investment: track indexation status, keyword rankings for the target terms, and whether internal link clicks are actually moving traffic between pages in the cluster, not just whether pages exist.

Where Founder Judgment Still Matters

Be honest about what this workflow is and isn't. It replaces the mechanical grind: keyword research spreadsheets, brief templates, first-draft writing, and manually checking which posts should link to which. It does not replace deciding which cluster to build first, whether a competitor's gap is worth chasing, or whether a specific claim in a draft actually reflects how your product works.

Read the full breakdown of what agentic SEO is and how AI agents actually run SEO workflows if you want the broader picture beyond content clusters specifically, including how the same agent-and-MCP pattern applies to technical audits and rank tracking.

Now for the uncomfortable part: even a perfectly drafted, perfectly linked cluster can fail if the underlying structure is wrong. Two pages competing for the same search intent, thin content padded to hit a word count, or answers buried three scrolls down instead of stated upfront, any of these can undermine months of otherwise solid work. The agent handles execution. You still own the strategy.