SEOLint is a technical SEO scanner with a Search Console feedback loop and a PR agent that ships the fixes. This guide gives you the order of operations, where to click first, and how to read what the agent shows you so the loop starts paying back in the first week.
The three-step loop
- Connect your site + Search Console + (optional) GitHub.
- Read Todos: deduped tasks with a copy-paste fix and an auto-close condition.
- Delegate to the PR agent or Claude Code via MCP.
Connect once, then let it run
Three things make the system work. Everything else is nice-to-have. Do them in this order and the first 15 minutes covers everything the loop needs.
Bind your site (30 sec)
Run your first scan at /scans. The first scan is on us. Every scan after that runs on your own Claude API key; we never pay for AI on your behalf. Free tier is 100 scans per day.
Connect Google Search Console (2 min)
The single biggest value multiplier in the whole product. Without it the scanner gives you best-practice advice. With it, every fix gets graded against real impressions, clicks, and position at 14, 28 and 90 days. The agent prioritises issues by traffic weight. Wire it up in /connections.
Install the GitHub App (3 min, optional)
Lets the PR agent open real pull requests for mechanical fixes. Without it you still get the full todo list and can do the fixes yourself; the bot just has nowhere to push changes. Sign off a Search Console-connected site here and the repo picker appears in the GitHub install flow.
That's the whole setup
The Todos tab is the product, trust it
Every scan becomes todos at /home/todos. One task per issue type across the whole domain, with the list of affected pages baked in. If twelve pages miss a meta description, you see one todo that says affects 12 pages, not twelve separate rows.
Each todo carries four things you need to act on it:
- What: the issue in the page's own context.
- Why: why it matters for this specific site, grounded in GSC traffic on the affected pages.
- Fix: concrete instructions you can paste straight into Claude or into code.
- Fixed when: the exact condition the agent watches for. When that holds on the next scan, the todo auto-closes.
Three actions sit on every row:
Copy as markdown: full task to clipboard, paste into Claude Code or a PR body.Create PR: only when the fingerprint is auto-fixable (see next section).Mark done/Won't do: status change. Won't do is domain-wide by default.
Domain-wide dismissals save repeat work
Let the PR agent ship the mechanical fixes
About a third of what a typical scan flags is a mechanical HTML-level fix. The PR agent ships a dedicated fixer for each. Click Create PR on a todo and the bot branches your repo, writes the file change (Haiku-generated copy when the fix needs text), opens a PR against the default branch, and attaches a summary.
The catalog today:
- Meta description
- Page title
- Canonical URL
- Open Graph tags
- Twitter card
- HTML lang attribute
- Viewport meta
- Organization JSON-LD
- sitemap.xml
- robots.txt
- llms.txt
- skills.md
Review the PR, merge it, move on. When the next scan picks up the merged fix, the todo auto-closes. Things with genuine subjective judgment (alt text on 200 images, rewriting a title for brand voice) stay with you.
Use Claude Code in the loop when you want it
The MCP server exposes SEOLint as a set of tools Claude Code (or Cursor) can call natively. The one to know first is next_issue:
# inside Claude Code, after adding seolint as an MCP server: "call seolint.next_issue for example.com and apply the fix"
Claude pulls the top todo back as markdown, opens the affected file in your workspace, applies the change, and commits. Combined with Claude Code's auto-PR flow you get from finding to shipping without leaving the terminal.
Week by week, what you should see
The loop compounds. If you've done the three setup steps and stopped, here's the rhythm.
Week 1: Baseline
First scan + GSC sync. Todos appear with the full issue breakdown. You ship the obvious fixes (or the PR bot does). The first recurrence-review todos can't fire yet; nothing's been dismissed long enough.
Week 2: First grading verdicts
14-day verdicts arrive via the Memory tab. You can see which of your week-1 fixes moved impressions, CTR, or position, and which landed flat. Flat fixes open fix_review todos with four likely reasons to check.
Week 3+: Steady state
Weekly rotation picks up new issues as your site evolves. Performance regressions, GSC query drops, schema opportunities on new pages. Each becomes a todo with auto-close conditions. The list stays short because old things self-clean.
Things that look quiet but aren't
Silence is a feature
Same with the weekly brief email. If the agent has nothing material to report (no new issues, no dismissed-and-returned fingerprints, no PRs shipped), you don't get an email. Silence by design.
Pricing, briefly
Free
$0/mo · no cardFull dev tool: scanner, memory, GSC integration, MCP read tools, every todo type. First scan is on us; bring your own Claude API key after. 100 scans/day.
SEOLint Agent
$99/mo · 7-day trialFree plus the autonomous layer: weekly auto-scans, next-action oracle, MCP write tools, GitHub PR bot, email agent. Still runs on your own Claude key. One site under agent watch, 200 pages tracked, 1,000 on-demand scans/mo.
Custom
multi-siteFor agencies and multi-site teams. Email support@seolint.dev.
Related reading
FAQ
How long until SEOLint pays for itself?
On a connected site with Search Console wired in, the first week gives you a deduped list of every technical SEO issue + a grading baseline. The PR agent can ship mechanical fixes immediately. Grading verdicts at 14 days tell you which fixes actually moved traffic.
Do I need the Agent plan to get value?
No. Free gives you the scanner, the Todos surface, memory per domain, GSC integration, and full MCP read tools. The Agent plan ($99/month, 7-day trial) adds weekly auto-scans, the PR bot, the next-action oracle, and MCP write tools.
What if the todos include things I don't want to fix?
Dismiss the todo domain-wide. The dismissal propagates to every page that has the same issue fingerprint. If a dismissed issue reappears in 3+ scans, the agent opens a recurrence-review todo asking whether the original decision still holds.
Can I use SEOLint from my editor instead of the dashboard?
Yes. The MCP server exposes scan, next_issue, open_issues, site_intelligence and get_scan as tools Claude Code or Cursor can call natively. Ask for next_issue and you get the top todo with its ready-to-paste fix, no context-switching.
Start running it
Free tier is the full dev tool. First scan is on us; bring your own Claude API key after that.