VisiblyAI MCP Server

SEO tools for Claude Code. Free local tools + paid API-powered analysis.

Production routing: MCP clients connect to https://mcp.visibly-ai.com/mcp. The installed package sends API-backed tool calls to https://app.visibly-ai.com/api/v1/mcp. The marketing domain visibly-ai.com is not an API host.

Free tools (no API key needed): keyword classifier, SEO checklists, best practices, URL analysis, Google guidelines, workflow skills.

Paid tools (require API key + credits): traffic analysis, keyword research, backlinks, competitors, OnPage SEO audit, PageSpeed/Core Web Vitals, SERP analysis, sitemap audit, structured data validation, hreflang checks, link checking, SEO agents, SEO workflows, advanced keyword classification.

Quick Start

Option 1: Remote server (zero install, recommended)

No Python or pip needed. Just add a URL to your Claude Code config:

# With API key (all 83 tools):
claude mcp add --transport http \
  --header "Authorization: Bearer lc_your_key" \
  visiblyai https://mcp.visibly-ai.com/mcp

# Without API key (8 free tools only):
claude mcp add --transport http visiblyai https://mcp.visibly-ai.com/mcp

Option 2: uvx (local, no install needed)

claude mcp add --transport stdio \
  --env VISIBLYAI_API_KEY=lc_your_key \
  visiblyai -- uvx visiblyai-mcp-server

Option 3: pip install (local)

pip install visiblyai-mcp-server
claude mcp add --transport stdio \
  --env VISIBLYAI_API_KEY=lc_your_key \
  visiblyai -- visiblyai-mcp-server

Then restart Claude Code.

No API key? Free tools work without one. Get an API key at app.visibly-ai.com/register to unlock paid tools.

Tools

Free (local, no credits) — 8 tools

Paid (API-powered, uses credits) — 20 tools

Google & Project (API key required, 0 credits) — 5 tools

These tools use your own Google OAuth tokens connected via the VisiblyAI platform.

Project data & content (API key required, 0 credits, read-only) — 15 tools

Nothing here writes back to visibly or spends credits. Team members read on the project owner's data; area roles apply (analytics / seo_tools / content).

Agent self-optimization skill

Load get_skill(name="content-nss-optimize"), then ask: "Write this article using the project briefing, improve it yourself with Visibly scoring until NSS 80, save the best draft and show its editor link." The portable SKILL.md works as instructions for Claude, Codex or another MCP-capable agent. It also supports a requested target of 70, checks headings/text/links, stops at a plateau, and re-scores the saved text. Visibly generation is never used for this loop. External agent tokens and separately requested data analysis retain their usual costs.

Claude Code and API-key-capable MCP clients use the existing HTTP connection. An OpenAI Responses API agent can use the same server_url and pass the Visibly key as authorization (official MCP guide). Direct ChatGPT plugin sign-in additionally needs an OAuth-compatible login; the current API-key connection alone is not a completed ChatGPT plugin integration (authentication requirements).

Content workflow (API key with scoped rights) — 35 tools

Read the editor context, research keywords, create and analyze a query, write and optimize drafts, approve articles, configure CMS connections, publish, and review optimizer suggestions. The Visibly server operator must enable the relevant write stage. Grant only the required rights in Visibly Settings → API key & MCP; existing keys do not receive new rights automatically.

Writing in Claude: read an existing analyzed query and briefing, let Claude write the text, then call submit_article_draft with that query_id and its current expected_query_revision. Show the returned editor_url as a clickable link. Saving uses 0 Visibly credits and never starts Visibly text generation. The editor automatically scores the text for 0 credits against the analyzed query; score_text also provides free feedback within Claude. New SERP/content analyses via create_content_query are separately paid. Without a ready analysis, full NSS scoring is unavailable. Do not call generation, regeneration, queueing or paid optimization simply to save or score Claude's text. Automatic browser opening depends on the Claude client; the MCP response provides the link.

Webhook secrets and cp_ keys created through MCP are returned once. Store them in the hosting provider's secret manager; never paste them into source code or logs. The Visibly MCP creates and tests the CMS connection but does not write Vercel, Railway, Hetzner, or other host environment variables. Use that provider's authorized MCP or secret manager for deployment configuration.

The paid optimizer run and article generation require the spending:execute key scope; the caller must still pass the project budget, tier and credit checks. The server reports their actual errors rather than silently degrading into free actions.

Limits: 60 requests/minute per key, responses up to 256 KiB (lists continue via page.next_offset), requests up to 512 KiB. Errors come as {"error": "<code>", "message": ...} (not_found, role_no_access, rate_limited, result_too_large, ...).

Examples

In Claude Code, just ask naturally:

> Classify these keywords: "seo tool kaufen", "was ist seo", "seo agentur berlin"

> Give me the blog SEO checklist in German

> What are the best practices for title tags?

> What does Google say about EEAT?

> Get the top keywords for example.com

> Run an OnPage SEO analysis on https://example.com/page for the keyword "seo tool"

> Run a full SEO performance audit for example.com (project_id: 12)

> Search the knowledge base for structured data best practices

Configuration

Environment Variable

Getting an API Key

1. Sign up at app.visibly-ai.com/register
2. Go to Settings > API Keys
3. Create a new key (starts with lc_)
4. Add it to your Claude Code MCP config

Subscription Tiers

Requirements

• Python 3.10+
• Claude Code CLI

Development

Run the test suite with PYTHONPATH=src, not a plain python -m pytest:

PYTHONPATH=src python -m pytest tests -q --ignore=tests/integration

Without PYTHONPATH=src, python -m pytest can resolve visiblyai_mcp against an installed package version instead of the local src/ checkout, so tests run against the wrong code. --ignore=tests/integration skips the tests that need a live API key.

License

MIT