Scan, Fix, Verify with AI

Use Sitepager and your AI assistant to find and fix issues on any website.

Sitepager scans your published site and identifies issues such as broken links, missing metadata, structural problems, and visual differences against the baseline. Your AI assistant reads these results, proposes fixes, and applies them to your site or codebase once you approve. After publishing the changes, you run the scan again to verify the issues are resolved.

This guide walks through the full workflow from scan to fix to verification.

Your assistant proposes. You approve. Nothing changes on your site without your review.

Quick answer

Connect the Sitepager MCP server to your AI assistant. Then follow this loop:

Run a Sitepager scan
Ask your assistant to review the issues
Ask your assistant for a fix plan
Review and approve the plan
Your assistant applies the fixes
Publish your changes
Run the scan again and verify

Requirements

Before starting, make sure you have:

A Sitepager account with a scan configured for your site
The Sitepager MCP server connected to your AI assistant
An AI assistant that supports MCP (Claude Code is recommended)

If you have not connected the Sitepager MCP yet: → Connect Sitepager to Your AI Assistant

How Sitepager and your AI assistant work together

Sitepager and your AI assistant handle different parts of the workflow.

Sitepager scans your published site and surfaces:

Broken internal and external links
Missing metadata or headings
Missing image alt text
Visual differences against the baseline
Pages added or removed since your last scan

Your AI assistant reads these results and proposes fixes. With a platform MCP such as Webflow MCP, or with access to your codebase, it can apply those fixes directly to your site.

Your assistant reads Sitepager scan results, proposes fixes, and applies approved changes. You remain in control of every step.

Sitepager Scan→AI Assistant→Plan & Approve↑↓Verify fixes←Publish←AI applies fixes

Sitepager identifies issues. Your AI assistant plans and applies fixes with your approval.

The workflow

Each step in the loop has a specific job. Here is the full sequence before we walk through it:

flowchart TD
  A[Run scan] --> B[Review issues]
  B --> C[Plan & approve fixes]
  C --> D[Apply fixes]
  D --> E[Publish]
  E --> F[Verify]
  F -->|Clean| G[Set new baseline]
  F -.->|Issues remain| B

Step 1: Run a scan

Run your scan from the Sitepager dashboard. Wait for it to complete.

You can also ask your assistant to trigger it:

Prompt

Run the Sitepager scan “[scan name]” and wait for it to complete.

Step 2: Review the issues

Ask your assistant to summarize what the scan found.

Prompt

Review the latest run of the Sitepager scan “[scan name]” and tell me the most important issues to fix.

Your assistant reads the results and returns a prioritized breakdown. It tells you what issues were found, which pages are affected, and which fixes will have the most impact.

Start with the highest-impact issues first.

Claude Code showing prioritized issue summary with breakdown by severity — Your assistant returns a prioritized breakdown of what to fix first.

Read through the full summary before moving on. This is your chance to understand what is wrong before anything is touched.

Step 3: Ask for a fix plan

Before fixing anything, ask your assistant to show you exactly what it will change.

Prompt

Show me a plan for fixing the issues. For each one, list the current value, what you propose to change it to, and which pages are affected.

Your assistant produces a detailed plan covering the current state, the proposed change, and where it will be applied.

Work through issues section by section. If the number of issues is small, reviewing everything in one plan is fine. For larger sites or scans with many issues, scope the plan to one issue type at a time. Ask for a plan, review and apply it, publish and verify, then move to the next. This keeps each review focused and makes it easier to catch anything unexpected.

Step 4: Approve and refine

Once you have reviewed the plan, confirm what you want your assistant to proceed with.

You can approve the plan as-is, adjust specific items before proceeding, or write the copy yourself and ask your assistant to apply your exact values. Your assistant will not make any changes until you give the go-ahead.

Step 5: Apply the fixes

Your assistant applies the approved fixes. Where the fixes land depends on how your site is built:

Platforms with an MCP server like Webflow. Fixes are applied directly to the site through the MCP. If your site is on Webflow, there are additional prompts and steps specific to the Webflow workflow. See Fix Webflow Issues with AI.
Codebase-managed sites like Next.js, Astro, React, static sites, and AI-maintained sites. Your assistant edits the files in your repo directly.

Once complete, your assistant confirms what was applied.

Step 6: Publish your changes

Review the changes before publishing. How you publish depends on your setup. You can deploy from your repo, publish from a CMS designer, or push your changes live however your team normally does.

Step 7: Run the scan again

Once your changes are live, run the scan again.

Prompt

Run the Sitepager scan “[scan name]” again and wait for it to complete.

Your assistant triggers a new run and waits for it to finish.

Step 8: Verify the fixes

Sitepager runs the comparison automatically. Your assistant reads the result and surfaces what was fixed, what remains, and whether anything new appeared.

Claude Code comparing scan results before and after fixes, showing resolved issues — Sitepager confirms what was fixed and what remains.

If issues remain, go back to step 3 and repeat for the next issue type.

Step 9: Set a new baseline

Once the scan comes back clean and you are happy with the result, update your baseline in the Sitepager dashboard. This makes the clean scan the new reference point for every future update.

Baseline updates are not currently available via MCP and need to be done from the dashboard.

→ Understanding Baselines

Sitepager dashboard showing a new baseline being set after a clean scan — Set the clean scan as your new baseline so future comparisons start from here.

Running the full loop in one prompt

If your site is maintained by an AI assistant with access to your codebase, such as Claude Code or Cursor, you can run the entire loop with a single prompt:

Prompt

Review the latest Sitepager scan, propose fixes, apply them to the codebase after I approve, then re-run the scan and confirm the issues are resolved.

Your assistant reads the scan, plans the fixes, pauses for approval, applies them, triggers the re-run, and verifies the result end to end.

Closing the loop

Once the fixes are applied, the follow-up scan verifies the result and confirms that the issues are resolved.

At this point the workflow is complete:

Scan → review → plan → approve → fix → publish → verify.

From here, the same loop becomes part of your normal workflow. Run a scan whenever you make updates, review the results, and fix issues before they reach visitors.

Over time this turns website maintenance into a repeatable process instead of manual spot-checking across pages.

Beta feedback

The AI + MCP workflow is currently in beta.

If something doesn’t behave as expected or you have ideas for improving the workflow, send us a note at support@sitepager.io or through the in-app chat. Feedback during beta helps us improve the automation and documentation.