Case Study: How We Built Seovane, an MCP-Native Organic-Growth Engine

Organic growth is no longer one game. A brand has to rank in Google, get cited in AI answers, and show up in the Reddit threads where people actually ask for recommendations. More and more teams already have an AI assistant they trust with real work, but that assistant cannot grow a business it has no tools to act on. It can advise on strategy; it cannot research keywords, write and fact-check a piece, or publish it where customers are looking.

We built Seovane to close that gap. This is not a story about a traffic number; it is a story about how the engine is built and why we built it the way we did. Seovane is an MCP server and dashboard: you connect it to your AI once, and the AI gains Seovane's entire organic-growth toolkit, keyword research, content generation, fact-checking and multi-surface publishing, so it can promote your business and bring you customers, with you watching from the dashboard or letting it run on its own.

The problem we set out to solve

The problem was never a lack of things to write about. It was that every stage lived in a different tool, and the AI that could orchestrate them had no way to actually touch them.

So the brief was to expose the whole organic-growth chain as tools an AI can call, where every stage is a deliberate step that runs with a human approving or fully on its own.

What we built: an MCP server your AI plugs into

The decision that shapes everything else is the interface. Instead of building yet another dashboard a person has to log into and operate, we exposed Seovane as an MCP server. Connect it once to your AI assistant or agent, and every capability below, keyword research, planning, drafting, fact-checking, publishing, becomes a tool the AI can call directly. The dashboard does not disappear; its job changes. It becomes where you watch what the AI is doing, approve actions, and read results, rather than where you do the work by hand.

This matters because it puts the growth toolkit inside the thing already doing your thinking. Your AI stops handing you advice it cannot act on and starts running the actual work, on your data, with your keys, under your approval.

The toolkit: from keyword to published

Behind the MCP interface, Seovane is a content pipeline that a topic moves through, stage by stage, with each stage recorded and exposed as a tool the AI can call. A keyword becomes a plan, the plan becomes researched angles, the angles become an outline, the outline becomes a draft written in a specific author voice, the draft is fact-checked and linked, and only then is it published. Nothing skips a step, which is exactly what keeps quality stable when the volume goes up.

Planning before writing

Most AI content tools start at "write me an article." Seovane gives the AI a stage earlier, on purpose. It builds a keyword universe, scores those keywords, and clusters them into topics, so the AI chooses what to write based on opportunity rather than whim. From the clusters it spawns a content plan, which is what makes the output a coherent body of work that builds topical authority, instead of a pile of disconnected posts.

Writing content that earns its ranking

A draft is only useful if it is trustworthy, so the generation stage is built around credibility, not word count. Content is written in a defined author persona with explicit E-E-A-T signals, so it reads as if a real, qualified person wrote it. Before anything is published, it passes a fact-checking stage and an internal-linking stage that ties it into the rest of the site, and media is attached. For international clients, the system handles hreflang so the same topic can exist correctly across languages. These are the unglamorous steps that separate content that ranks from content that gets ignored.

Publishing across every surface, your way

Ranking in Google is only one third of the job, so Seovane publishes to more than one place. It pushes finished articles to the client's own sites, and it runs a dedicated Reddit engine for the conversations where buying decisions actually happen. The Reddit engine is built carefully: it authenticates over OAuth, supports a bring-your-own Reddit app per workspace so each client posts under their own identity, and offers two modes. In co-pilot mode a human approves every action, often straight from a Telegram bot or the dashboard; in automated mode the AI runs the cadence itself. We built both modes because trust in automation is earned gradually, and the system should meet a client wherever they are on that curve.

The dashboard and the architecture behind it

The dashboard is the human half of the system: it is where you connect the MCP server, hold your API keys, approve what the AI proposes, and see what has been published and how it performed. Everything the AI does through MCP shows up there, so handing work to an agent never means losing sight of it.

Built multi-tenant, with secrets done right

Seovane is multi-tenant by design, with strict per-workspace isolation through row-level security on every query. Because the engine acts on each client's behalf across third-party platforms, and because an AI is calling those tools, it has to hold credentials safely, so we built an encrypted secrets vault with tightly scoped write policies and credential-column lockdown, and each workspace brings its own API keys. The whole platform went through repeated security-hardening and audit passes, including a drift reconciler that checks the live database against what the migrations say it should be, because a system that holds other people's keys and lets an AI post under their names has to be trustworthy first.

Under the hood

Seovane is a custom application, not a pile of SaaS subscriptions wired together.

• An MCP server as the primary interface, so any connected AI assistant or agent can call Seovane's tools directly, with a dashboard for connecting, approving and monitoring.
• Next.js and PostgreSQL (Supabase), multi-tenant with row-level security, scheduled with pg_cron so the pipeline can run on its own.
• A keyword engine that builds a scored keyword universe, clusters it into topics, and spawns a content plan, so the AI writes by opportunity, not by guess.
• A staged content pipeline: research, angles, outlines, drafting in an E-E-A-T author persona, fact-checking, internal linking, media, repurposing, and a learn loop that feeds results back into planning.
• A multi-surface publisher that ships to client sites and runs a dedicated Reddit engine with OAuth, a bring-your-own Reddit app per workspace, and co-pilot or automated modes.
• DataForSEO for live SERP and keyword data, with Reddit-thread discovery driven from search rather than scraping.
• An encrypted secrets vault with scoped write policies and credential lockdown, plus Telegram approvals and notifications.
• Audit-grade hardening: repeated security passes and a drift verifier that reconciles the live schema against the migrations.

Why weeks, not quarters

A platform this broad usually sounds like a year of work. It is not, because the build follows the same process we use on every project: a tight scope, an agent-assisted build, and a deployment that is monitored from day one. We break the method down in how we ship custom apps in weeks, and the build-versus-buy logic in build vs buy: when a custom app wins.

What we would tell anyone considering this

A few honest lessons from the build.

• Give the AI tools, not just advice. The leap is not a smarter prompt, it is an interface. Once your AI can call real tools over MCP, it stops describing what you should do and starts doing it.
• Credibility is a pipeline stage, not a prompt. Author personas, fact-checking and internal linking are explicit steps because that is what makes content rank and survive, rather than read as generic filler.
• Earn the right to automate. Co-pilot mode first, automated mode second. Letting an AI hold a client's keys and post under their name means trust has to be built deliberately, in the architecture as much as the workflow.

If you want your AI to actually grow your business across Google, AI answers and Reddit, instead of just advising you on it, talk to us. We will look at how your content gets made today and show you what an MCP-connected engine would change before we build anything.