Visibly AI — SEO MCP Server

Claude Code — SEO Agency & Freelancer Starter

A batteries-included Claude Code blueprint for SEO freelancers and agencies. Wire up the Visibly AI MCP, get ready-made slash commands for the full client workflow — Status-Quo → Potential Analysis → Offer → CI-compliant PDF — and a smart hook that nudges Claude toward real SEO data instead of guessing.

Built and battle-tested by [Antonio Blago](https://www.antonioblago.com) — SEO Freelancer & creator of the Neuro-SEO System®.

Quickstart • What's inside • Workflows • CTR Model • Customise • Get a free key • Book a call

---

Why this exists

SEO consulting is the same five jobs over and over: pull the real ranking data, map it against the client's keyword set, quantify the opportunity, turn that into an offer, and ship a presentable PDF. Claude Code can do all of it — *if* you give it the data sources, the methodology, and the guardrails.

This repo packages exactly that — and it runs without any API key. Install it and the bundled .mcp.json connects to the Visibly AI MCP keyless, giving you the free knowledge tools (keyword classification, SEO checklists, Google guidelines, URL-structure analysis) plus the full local methodology; bring your own Search Console export and the entire Status-Quo → Potential → Offer → PDF chain runs offline. Add a Visibly AI key when you want the data engine on tap — live GSC/GA at 0 credits, keywords, backlinks, competitors and on-page audits through one MCP. Free by default, pro when you need it.

✨ What's inside

🚀 Quickstart

There are two ways to use this — pick one.

Option A — Install as a plugin (fastest)

/plugin marketplace add AntonioBlago/claude-code-seo-starter

/plugin install seo-starter@antonioblago

You get the /visibly-seo-status-quo, /visibly-seo-potential, /visibly-seo-offer, /visibly-seo-pdf-build commands, the auto-invoked SEO skills, the SEO-intent hook, and the keyless Visibly AI MCP — wired in. No key needed to start — approve the MCP server and you have the free knowledge tools plus the full local workflow.

Want the full data engine (live GSC, keywords, backlinks, on-page)? Add a Visibly AI key — grab one at visibly-ai.com (~30s, no card), then connect it once (resolves at write time, stays out of git):

claude mcp add --transport http visiblyai https://mcp.visibly-ai.com/mcp --header "Authorization: Bearer lc_xxxxxxxxxxxxxxxx"

Restart with /reload-plugins if needed. See docs/setup.md for the pro tier, Google-OAuth (0 credits) and optional DataForSEO wiring.

Option B — Clone the template

Use the full repo (docs, PDF template, setup script) as a project scaffold.

1. Prerequisites

• Claude Code installed with the native installer: curl -fsSL https://claude.ai/install.sh | bash (macOS, Linux, WSL) or irm https://claude.ai/install.ps1 | iex (Windows PowerShell). New to Claude Code? Start with claude-homebase: /homebase sets up your global CLAUDE.md and project structure, then pick the SEO path to land here.
• *(Optional)* a Visibly AI account for the pro tier — [sign up free](https://visibly-ai.com). Not required: the starter runs keyless.
• Python 3.11+ (only needed for the PDF / data-crunching helpers)

2. Get the blueprint

# Use it as a GitHub template, or just clone:

git clone https://github.com/AntonioBlago/claude-code-seo-starter.git cd claude-code-seo-starter

3. (Optional) add your Visibly AI key

Skip this for the free tier — the keyless .mcp.json already works. To unlock the full data engine, create a free account at [visibly-ai.com](https://visibly-ai.com), copy your lc_... key, and connect it once (resolves at write time, stays out of git):

claude mcp add --transport http visiblyai https://mcp.visibly-ai.com/mcp --header "Authorization: Bearer lc_xxxxxxxxxxxxxxxx"

Full tier breakdown, Google-OAuth (live GSC/GA at 0 credits) and optional DataForSEO wiring: [`docs/setup.md`](docs/setup.md).

4. Launch

claude

On first run, Claude Code will ask you to approve the Visibly AI MCP server. Approve it, then try:

/visibly-seo-status-quo example.com

That's it. On the free tier, feed a GSC export to the Python templates; with a key, data is pulled live. Either way you're running the full SEO workflow.

🔁 The workflows

Each slash command corresponds to one phase of a real client engagement. They chain:

/visibly-seo-status-quo <domain>   →   /visibly-seo-potential <domain>   →   /visibly-seo-offer <domain>   →   /visibly-seo-pdf-build <script.py>

(where do we (what's the upside, (what should the (hand the client a rank today?) in clicks & €?) client buy?) polished PDF)

Full methodology is documented in [`docs/workflows.md`](docs/workflows.md).

📈 The CTR model

The potential analysis is only as good as its click-through-rate curve. This repo ships the [Keyword Study 2026](https://www.antonioblago.com/keyword-study-2026-organic-search-ctr) — CTR by position from first-party Google Search Console data (1.3M keywords, 94 domains) — instead of the inflated textbook numbers most templates still use.

The differentiator: the curve is intent-aware. A navigational keyword at position 1 earns ~8.9% CTR; an informational one ~3.2%. Forecasting with one blended number quietly mis-states the upside. See [`docs/ctr-model.md`](docs/ctr-model.md) for the full intent-by-position table and the Python to apply it. Honest inputs → defensible forecasts → offers that survive scrutiny.

🔍 Transparency — what runs where

No surprises about data flow:

• Keyless (default): .mcp.json connects to mcp.visibly-ai.com without an account. Only the free knowledge tools respond (checklists, guidance, Google guidelines, keyword classification, URL-structure checks) — no client data involved.
• With a key: the data tools (query_search_console, query_analytics, keywords, backlinks, on-page) route through the Visibly AI backend, hosted on Railway in the EU. Your GSC/GA data passes through that backend to reach Claude — that's the product, not a side effect. And it takes three explicit opt-ins before a single GSC row can flow: create a Visibly account, add your API key, and connect your Google account via OAuth inside Visibly. Without the Google consent, query_search_console has no property to read. Not acceptable for a client? Don't add the key: feed a GSC export to the local Python templates instead — the full workflow still runs.
• Permissions: .claude/settings.json auto-allows only metadata and the free knowledge tools. Every tool that reads client data (GSC, GA, keywords, backlinks, on-page, crawling) sits in ask — Claude prompts before using it.
• The hook (.claude/hooks/seo-check.sh) runs locally, sends nothing anywhere, and only injects a reminder to use real data instead of guessing when your prompt looks SEO-related. Don't want the nudge? Delete the hook.

🎨 Customise it

Everything is plain Markdown and shell — fork and adapt:

• Branding — edit CLAUDE.md and templates/pdf_example.py with your colours, fonts, contact block.
• Data source — keyless by default; add a Visibly key for the full engine, or wire your own (GSC export → Python templates, DataForSEO, GA4, Ahrefs…) in .mcp.json. See docs/setup.md. Free walkthrough for building your own GSC access from scratch: Indexing Checker für die Google Search Console (German).
• Workflows — the .claude/commands/*.md files *are* the workflows. Rewrite them in your own words.
• Guardrails — extend .claude/hooks/seo-check.sh with your own keyword triggers.

See [`docs/setup.md`](docs/setup.md) for a deeper walkthrough and [`docs/best-practices.md`](docs/best-practices.md) for the hard-won lessons.

📂 Repo layout

claude-code-seo-starter/

├── .claude-plugin/ # makes this repo an installable plugin + marketplace │ ├── plugin.json # plugin manifest │ └── marketplace.json # marketplace catalog (one repo = both) ├── .mcp.json # Visibly AI MCP connection — keyless (free tools); add key for pro ├── server.json # MCP-registry manifest (publish via mcp-publisher) ├── .env.example # copy → .env — optional key + separate data sources ├── CLAUDE.md # project instructions Claude reads on every run ├── .claude/ │ ├── commands/ # /visibly-seo-status-quo /visibly-seo-potential /visibly-seo-offer /visibly-seo-pdf-build │ ├── skills/ # auto-invoked SEO skills (status-quo, potential, offer, pdf) │ ├── hooks/ # seo-check.sh + hooks.json — SEO-intent nudge │ └── settings.json # hook registration + permissions (free tools allowed, data tools ask) ├── docs/ │ ├── setup.md # detailed setup │ ├── workflows.md # the 4-phase methodology │ ├── ctr-model.md # empirical CTR curve │ ├── folder-structure.md # client knowledge base + CI base conventions │ ├── best-practices.md # lessons learned │ └── promote.md # discoverability checklist (registry, awesome lists) ├── templates/ │ ├── ci/ # CI base — brand.py (constants) + CI.md (reference) │ ├── client-template/ # copy → clients/ / (CLAUDE.md + _knowledge/) │ ├── status-quo-template.md # fill-in status-quo skeleton (classification, quick wins) │ ├── potential-template.md # fill-in potential skeleton (deltas, clusters, ROI scenarios) │ ├── offer-template.md # fill-in offer skeleton (phases, pricing, ROI, terms, CTA) │ └── pdf_example.py # minimal CI-PDF starter (fpdf2) └── scripts/ └── check-setup.sh # verify your env is wired correctly

Related repos

| [claude-context-guard](https://github.com/AntonioBlago/claude-context-gu