What Claude gets access to
Once connected, Claude has three new tools available in every conversation:
| Tool | What it does |
|---|---|
scan_website | Scan a URL: returns structured issues with severity and fix prompts |
get_scan | Fetch a previous scan result by ID |
list_scans | List recent scans for your account |
Step 1: Get your API key
SEOLint issued an API key on signup, generated instantly on the dashboard and shaped like sl_live_xxxxxxxxxxxx. Keys are no longer issued: the product shut down in July 2026. The MCP setup below is kept because the pattern transfers to any MCP server.
Step 2: Add to Claude Code
Run this once in your terminal. Replace sl_YOUR_KEY_HERE with your API key from step 1:
claude mcp add seolint \ --env SEOLINT_API_KEY=sl_YOUR_KEY_HERE \ -- npx -y seolint-mcp
You should see Added stdio MCP server seolint in the output.
Important: Start a new Claude Codesession after adding. The MCP server only loads on startup. After that, it's available in every session.
Or use the CLI directly
If you just want a quick scan without the MCP setup:
SEOLINT_API_KEY=sl_YOUR_KEY_HERE \ npx @randomcode-seolint/cli scan https://mysite.com
Returns a markdown report you can paste into Claude or Cursor. Add --json for structured output.
Step 3: Ask Claude to scan
Open a new conversation and ask Claude naturally. Some examples:
“Scan https://mysite.com and tell me what SEO issues to fix.”
“Scan my homepage and apply the fixes to the codebase.”
“What changed in the SEO score since last week? Check the last two scans.”
“Scan https://mysite.com/blog/my-post and give me the H1 and title tag fix.”
When Claude Code has access to your repo, it can apply the fixes directly, not just describe them.
What Claude sees
Each issue in the scan result includes a fix_prompt field. Each one is a ready-to-use LLM instruction Claude can execute immediately:
{
"id": "missing-h1",
"severity": "critical",
"title": "Missing H1 tag",
"fix_prompt": "Add a single <h1> tag to this page containing the primary keyword. The page is about [topic]. Place it as the first heading inside the <main> element."
}Claude reads this structure natively and can execute the fix in Cursor or Claude Code without any additional prompting from you.
FAQ
What is an MCP server?
MCP (Model Context Protocol) is an open standard that lets AI assistants like Claude connect to external tools and data sources. When you add SEOLint as an MCP server, Claude gains the ability to call scan_website() as a native tool, no copy-pasting URLs into a browser needed.
Does this work with Claude Code as well as Claude Desktop?
Yes. Claude Code supports MCP servers via the `claude mcp add` command. Once added, you can ask Claude Code to scan your site from the terminal or from within a Cursor session.
What plan do I need?
The SEOLint plan ($79/month) includes MCP server access plus the weekly agent watch, 200 AI credits/month bundled in, Search Console grading, and AI fix prompts. One site under agent watch, BYOK to skip the credit cap entirely.
Can Claude automatically apply the SEO fixes it finds?
Yes. When Claude has access to your codebase (via Claude Code or Cursor), it can read the scan results and apply the fixes directly. Each issue in the API response includes a structured fix_prompt field designed for this workflow.
Related reading
- Claude as Your Persistent SEO Agent
- Run SEO Audits From the Terminal | SEOLint CLI
- Automate SEO Audits with an AI Agent
- SEO checklist for developers
Official reference: Model Context Protocol documentation
SEOLint is no longer running
This article was written while building SEOLint, an SEO agent that scanned your site, opened the pull request, and graded each fix against Search Console. The product shut down in July 2026. The advice above still stands.
Read the post-mortem