Agentic-SEO-Skill MCP — 사용 사례, 설치 & 라이브 데모

왜 쓰나요

핵심 기능

16 sub-skills: technical SEO, content quality, schema, performance, links, GEO/AEO, sitemap, hreflang, strategy
10 specialist agents (technical, content, performance, schema, sitemap, visual, GitHub, benchmarking, data, verification)
33 evidence-collector scripts (32 Python + 1 shell) for auditable findings
Runs on Antigravity (project-local), Claude Code (user-global), and Codex (user-global)
Enforces current Google signals: INP not FID, full E-E-A-T treatment

라이브 데모

실제 사용 모습

agentic-seo-skill.replay ▶ 준비됨

0/0

설치

클라이언트 선택

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "agentic-seo-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/Bhanunamikaze/Agentic-SEO-Skill",
        "~/.claude/skills/Agentic-SEO-Skill"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop → Settings → Developer → Edit Config 열기. 저장 후 앱 재시작.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "agentic-seo-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/Bhanunamikaze/Agentic-SEO-Skill",
        "~/.claude/skills/Agentic-SEO-Skill"
      ],
      "_inferred": true
    }
  }
}

Cursor는 Claude Desktop과 동일한 mcpServers 스키마 사용. 프로젝트 설정이 전역보다 우선.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "agentic-seo-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/Bhanunamikaze/Agentic-SEO-Skill",
        "~/.claude/skills/Agentic-SEO-Skill"
      ],
      "_inferred": true
    }
  }
}

Cline 사이드바의 MCP Servers 아이콘 클릭 후 "Edit Configuration" 선택.

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "agentic-seo-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/Bhanunamikaze/Agentic-SEO-Skill",
        "~/.claude/skills/Agentic-SEO-Skill"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop과 같은 형식. Windsurf 재시작 후 적용.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "agentic-seo-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/Bhanunamikaze/Agentic-SEO-Skill",
        "~/.claude/skills/Agentic-SEO-Skill"
      ]
    }
  ]
}

Continue는 맵이 아닌 서버 오브젝트 배열 사용.

~/.config/zed/settings.json

{
  "context_servers": {
    "agentic-seo-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/Bhanunamikaze/Agentic-SEO-Skill",
          "~/.claude/skills/Agentic-SEO-Skill"
        ]
      }
    }
  }
}

context_servers에 추가. 저장 시 Zed가 핫 리로드.

claude mcp add agentic-seo-skill -- git clone https://github.com/Bhanunamikaze/Agentic-SEO-Skill ~/.claude/skills/Agentic-SEO-Skill

한 줄 명령. claude mcp list로 확인, claude mcp remove로 제거.

사용 사례

실전 활용법: Agentic-SEO-Skill

Run an evidence-backed SEO audit with confidence labels

👤 SEO analysts tired of 'trust me' recommendations ⏱ ~60 min intermediate

언제 쓸까: You want findings you can defend to a skeptical engineering lead.

흐름

Trigger the full audit

Use agentic-seo-skill on https://site.com — full evidence-backed audit.✓ 복사됨

→ Scripts collect raw data (meta tags, schema, link graph, CWV); agents analyze
Review findings with proofs

Show the top 10 issues with their supporting evidence and confidence labels.✓ 복사됨

→ Each finding has a literal quote/data point behind it
Prioritized action plan

Order by impact × effort.✓ 복사됨

→ Ranked plan with numeric rationale

결과: An audit where every finding is citable.

함정

Low-confidence findings treated as high priority — Use the confidence labels — skip 'low' until high/medium are done

함께 쓰기: firecrawl

Audit a GitHub-hosted docs site with the GitHub-analyst agent

👤 Dev-tool teams with docs on GitHub Pages ⏱ ~40 min intermediate

언제 쓸까: Your docs live in a repo and you want a SEO audit tied to the repo structure.

흐름

Point the GitHub agent at the repo

agentic-seo-skill — audit github.com/acme/docs site. Output GITHUB-SEO-REPORT.md.✓ 복사됨

→ Report in the repo format with per-file recommendations
Open a PR with fixes

Turn the high-confidence findings into a PR.✓ 복사됨

→ PR with concrete diffs

결과: A SEO-improved docs site with a trackable PR.

함정

PR touches too many files at once — Split by finding type (meta vs content vs schema)

함께 쓰기: github

Optimize pages for Perplexity / ChatGPT / AI Overviews citations

👤 Content teams losing clicks to AI summaries ⏱ ~30 min intermediate

언제 쓸까: You want to be the cited source, not just another organic result.

흐름

Run the GEO/AEO sub-skill

agentic-seo-skill — GEO audit on https://site.com/post.✓ 복사됨

→ Findings tied to snippet-friendliness, entity clarity, citation signals
Apply and verify

Apply recommendations and re-verify.✓ 복사됨

→ Score deltas with evidence

결과: Pages restructured for AI-citation pickup.

함정

Over-optimization hurts human readability — The content-quality agent catches robotic-sounding edits

조합

다른 MCP와 조합해 10배 효율

agentic-seo-skill + firecrawl

Firecrawl does the JS-rendered crawl; agentic-seo interprets

Crawl with firecrawl, pipe rendered HTML into agentic-seo-skill for analysis.✓ 복사됨

agentic-seo-skill + github

Open PRs directly from audit findings

For the high-confidence findings, open a PR in the repo.✓ 복사됨

agentic-seo-skill + claude-seo-skill

Use the /seo slash commands for fast runs, agentic-seo for evidence-backed depth

First do /seo audit for the quick view, then agentic-seo-skill for the deep defensible audit.✓ 복사됨

도구

이 MCP가 노출하는 것

도구	입력	언제 호출
Technical SEO agent	URL	Baseline tech audit
Schema agent	URL	Structured data work
Performance agent	URL	CWV review
GitHub-analyst agent	repo URL	Audits on GitHub-hosted sites
Verification agent	prior findings	Before publishing audit

비용 및 제한

운영 비용

API 쿼터: None for the skill
호출당 토큰: 10-30k for full audit — evidence collection uses scripts, not tokens, for the raw data
금액: Free — skill is local
팁: Utility scripts are Python — run them as preflight outside the LLM loop to save tokens.

보안

권한, 시크릿, 파급범위

자격 증명 저장: No credentials — scripts hit public URLs only

데이터 외부 송신: Only to the sites you audit

Utility scripts run on your machine — review before running on sensitive internal URLs.

문제 해결

자주 발생하는 오류와 해결

Utility scripts fail to run

Check Python version and required packages; install dependencies from the skill's requirements file.

확인: python --version

Confidence labels all 'low'

Scripts couldn't collect enough raw evidence — check network access and JS rendering.

GitHub-analyst can't access repo

Set a PAT for private repos; public repos should work unauthenticated.

대안

Agentic-SEO-Skill 다른 것과 비교

대안	언제 쓰나	단점/장점
claude-seo-skill	You want slash-command UX and enterprise reporting	Claude-only; no multi-agent evidence framework
seo-geo-claude-skill	You want a phase-based lighter library	Less evidence-first methodology

Agentic-SEO-Skill