MCP Driving 30% of Signups

TL;DR

I personally built and shipped the first Model Context Protocol (MCP) server in the SEO/search space. After a weak-retention V1, I re-architected it from a local, subscription-gated install to a remote, pay-as-you-go server. It now drives ~30% of company signups — SEO professionals querying live keyword, backlink and SERP data in natural language from Claude, ChatGPT, Gemini, or any MCP-compatible client. I built it hands-on — including new API endpoints behind it — and use Claude Code daily for the PM work around it.

Context

AI agents needed an easy way to access SEO data. Model Context Protocol (MCP) was emerging as the standard for connecting AI assistants to external data sources. Being first to market could capture a brand-new segment — and I wanted SE Ranking to be the default SEO data source for every AI assistant.

Problem

• AI agents couldn't easily access live SEO data
• The technical barrier was high for non-developers
• The traditional Data API required coding knowledge
• SEO professionals wanted natural-language access to keyword, backlink and SERP data

What I Did

V1: Local Setup + Subscription (Weak Retention)

• Shipped an MCP integration that ran locally
• Required technical setup (Node.js, config files)
• Gated behind a subscription

What went wrong:

• Setup abandonment: most users abandoned during the Node.js install step. I was effectively asking SEO professionals to become developers.
• Wrong pricing model: a monthly subscription made sense for heavy users, but most people wanted occasional queries. The commitment felt too high for "let me just try this."
• Support burden: the few users who finished setup generated constant tickets about config issues, breaking updates, and localhost problems.

The signal I almost missed: the users who got it running loved it. The product worked — the delivery mechanism didn't. That insight drove V2.

V2: Remote + Pay-as-You-Go

• Re-architected to a remote-hosted MCP server
• Switched to pay-as-you-go (no subscription required)
• One-step setup in any MCP client

Result: setup dropped from a Node.js install to ~30 seconds. Same engine, new packaging.

claude mcp add --transport http seranking-mcp https://api.seranking.com/mcp

Authenticate once with Google OAuth and query live SEO data in natural language. Because MCP is an open standard, the same setup works in any MCP-compatible client — Claude, ChatGPT, Gemini, and beyond.

Key Decisions

Bet Early on an Emerging Protocol

I committed to MCP before it was widely adopted. Being first established SE Ranking as the go-to SEO data source for AI assistants.

Remove Friction at Every Step

V1 failed on friction. I removed setup complexity and the subscription commitment, so the first query is reachable in seconds.

Target Non-Developers

I realized the market wasn't AI developers (they use the Data API directly) but SEO professionals who wanted natural-language access. I designed V2 for them.

Technical Details

I wasn't only making the product calls — I was hands-on building it: I built and integrated new API endpoints behind the MCP and worked through the implementation to make the server run end to end.

• Built & integrated new endpoints: I built and wired up API endpoints behind the MCP and worked alongside engineering to make it run end to end — not just specced it.
• An open standard, not one-off integrations: MCP is a standard, so the server works with any MCP-compatible client (Claude, ChatGPT, Gemini, and any other) — one server, every client, instead of a separate integration per platform.
• Remote + Google OAuth to kill the install step: I moved hosting server-side with Google OAuth in front, so there's no Node.js install, no config file, and nothing to keep updated — the barrier dropped to a single authenticate-once flow.
• Pay-as-you-go metering on the existing Stripe wallet: instead of building new billing, I had usage-based metering ride the existing Stripe wallet — users top up and pay for what they query.
• Natural-language interface: I tuned the tool surface and responses for conversational queries against live keyword, backlink and SERP data.

Results

• SEO professionals querying live data in natural language on ChatGPT, Claude, and Gemini
• Opened the AI-agent and no-code markets (n8n, Make)
• Created a new acquisition channel independent of traditional marketing

Try it live

Everything you need to set it up — the live endpoint (https://api.seranking.com/mcp), the one-line install, and the Google OAuth flow — is on the SE Ranking MCP integration page.

Lessons Learned

1. Fail fast, iterate faster: V1 was wrong, but I learned in weeks what might have taken months of planning to hypothesize. The cost was low; the learnings were invaluable.
2. Friction kills products: same core engine, completely different outcomes. V1→V2 was a packaging change, not a technical rebuild. Sometimes the product isn't the problem — the delivery is.
3. Listen to the users who succeed: the signal was in my happiest users, not my churned ones. Those who completed V1 setup loved the product. That told me what to protect in V2.
4. First-mover advantage is real: being first to ship MCP in our space made SE Ranking the default choice — but only because V2 made it easy to adopt.
5. Find the real user: I thought I was building for developers but found product-market fit with SEO professionals. The best insight came from watching who actually used it, not who I designed it for.