Why - We already ship two MCP engines (one scores a page for AEO/GEO, one tracks what LLMs cite). Running them page by page and triaging into a to-do list was the tedious part. The magic button is the orchestration layer that emits one artifact an agent can run.
How it works (two-step flow)
- Run the skill → get a plan. Crawls every page, runs the full audit, ranks fixes by estimated score lift, folds cross-page gaps (no sitemap, missing llms.txt, schema absent sitewide) into single items. Output:
plan.json + plan.md.
- Run your agent against the plan. Each item carries a baked tool call + acceptance check, so the agent executes sequentially. Analyze-and-propose only - nothing changes until you run
apply.
Install (all three surfaces)
# Option A - Claude Code plugin (recommended)
/plugin install https://github.com/AutomateLab-tech/ai-seo-magic-button
# Option B - MCP server in any host (Cursor, Cline, Windsurf, Claude Desktop, ...)
npx -y ai-seo-magic-button mcp
# Option C - Claude skill only: copy skill/ into .claude/skills/ai-seo-magic-button/
{"mcpServers":{"ai-seo-magic-button":{"command":"npx","args":["-y","ai-seo-magic-button","mcp"]}}}
Example - Ask your agent: "Run the AI SEO magic button on example.com and show me the top 5 fixes." The generate_seo_plan tool returns the plan as structured content. CLI adds audit, apply (idempotent), verify (before/after score diff), sink (push items to a Notion board).
Links - Product page: https://automatelab.tech/products/skills-and-plugins/ai-seo-magic-button/ · GitHub: https://github.com/AutomateLab-tech/ai-seo-magic-button · npm: https://www.npmjs.com/package/ai-seo-magic-button · Plan format: https://github.com/AutomateLab-tech/ai-seo-magic-button/blob/main/docs/plan-format.md