Give every finding
a paper trail.

Crawl your site, see the affected URLs and keep the evidence. Run it from your terminal or connect the same engine to your coding agent.

Preview agent prompt
Act as my SEO audit agent using SEO Crawl Audit. Respond in the language I use in this conversation. Start from the current project when one is accessible: resolve its website URL from project evidence before asking me for details.

1. Resolve the project and adapt to your actual client and capabilities.
Use an explicitly supplied target URL first. Otherwise inspect the current working directory and applicable project instructions, SEO audit configuration, README, deployment configuration and public site URL/canonical settings. In a multi-site repository, use the active app only when it is unambiguous. Do not derive a domain from a folder name or Git remote, use unrelated examples/dependencies, or expose secrets. If exactly one intended website is supported by project evidence, state its URL, workspace and evidence, then proceed without asking for confirmation. Ask only for missing or ambiguous details (URL, app or environment); if no project is accessible, ask for the target URL before making website requests.
- With project filesystem access (Codex, Claude Code in CLI or the Claude app's Code tab, or another capable session): follow applicable AGENTS.md, CLAUDE.md and project instructions, then use the installed skill and connected MCP tools when available.
- With only connected tools: use the SEO Crawl Audit MCP connection and its configured workspace. Inspect only what the session actually exposes; do not assume terminal or filesystem access from an application name. Claude's Chat, Cowork and Code modes can expose different capabilities.
Discover the tools and read the available SEO Crawl Audit skill instructions before using them. Do not claim an integration is installed without checking. Keep artifacts inside the current project workspace. Check the MCP bound workspace before calling any audit tool: a binding to another project is not evidence of the target site. On a mismatch, use the CLI in the current project if available; otherwise explain the required reconfiguration. Do not scan or write artifacts in the other project. In a tools-only session, use the connection's configured workspace and ask only for details that cannot be determined there.

2. Plan, then audit.
Prefer the native MCP workflow: discover seo_audit_plan, seo_audit_scan, seo_audit_issues and seo_audit_report (client prefixes may differ). Inspect the crawl plan, start with a 100-page budget, respect robots.txt, and use the returned artifact paths. Page through the issue results rather than assuming one response contains every finding. If no native integration is available but a terminal is, use this fallback after replacing the URL:
npx seo-crawl-audit@0.17.5 scan <WEBSITE_URL> --pages 100
If neither route is available, explain the missing connection and point me to https://melnixdev.github.io/seo-crawl-audit/plugins/ instead of pretending to run an audit.

3. Investigate the evidence.
Review collected metadata, canonicals, indexability signals, internal links, hreflang and structured-data findings. Report affected URLs and evidence, separate confirmed problems from items needing manual investigation, and state checked-page counts, crawl limits and inaccessible pages. Do not claim full coverage, Google indexing or rankings from a bounded crawl.

4. Deliver an actionable result.
Link the local HTML report and JSON snapshot using the artifact access your client actually supports. Summarize the most important findings, then propose a prioritized fix plan with owners, affected pages and verification steps. If comparable saved runs exist, use the available comparison/history tools to explain changes without needlessly crawling again. Keep credentials and personal data out of shared outputs.

This request is for an audit and plan. Do not edit or deploy the website as part of it.
npx seo-crawl-audit@0.17.5 scan https://example.com/ --pages 100

Node.js 20.19+ required. Free and open source.

SEO Crawl Audit's coral crab, raising its claws beneath scan brackets

Open the report.
Follow the URL.

Start with crawl coverage, then move into pages and findings. Every issue includes evidence, its affected URL and a next step you can review.

SEO Crawl Audit Page Explorer rendered from deterministic four-page reference data
Actual 0.17.2 report interface rendered from four example pages and eight synthetic historical runs. These are reference data, not live measurements of quotes.toscrape.com. Open the same fixture report. No AI-client interface is simulated.

A useful audit leaves a trail.

  1. Plan a bounded crawlDiscover robots.txt and sitemap URLs. Choose a page budget before making requests.
  2. Inspect the evidenceReview titles, metadata, canonicals, links, hreflang and structured-data findings by URL.
  3. Check the next deploymentSave JSON snapshots and compare new, ongoing and resolved findings with comparable coverage.

Keep the details.
Share the HTML.

Filter findings by severity, rule or owner. Open a page to inspect its collected metadata and links. Keep the snapshot for another check, or share the generated report with your team.

A report is a local artifact you can reopen. Reading it does not require an SEO service account.

Issue explorer showing evidence and remediation in the deterministic reference report
The reference data intentionally removes metadata and repeats titles to demonstrate findings.
Report overview showing the deterministic reference data and synthetic issue history
Counts and historical dates are fixture inputs, not current site measurements or search-engine indexing evidence.

The same engine.
Your own workflow.

Use the CLI for a one-off investigation, a GitHub Action for regression checks or eight MCP tools inside your agent workflow.

  • Versioned JSON snapshots for repeatable comparisons
  • HTTP crawling by default; optional Playwright rendering
  • Explicit workspace boundaries for local MCP artifacts
Choose a client integration

See a report
in motion.

A short, silent preview assembled from real browser captures: explore the overview, pages, site metrics and issues.

This capture sequence uses deterministic reference data with synthetic history. It demonstrates report behavior, not Codex or Claude installation.

Download the report walkthrough or explore the fixture report.

Local by design.
Clear about limits.

The crawler runs on your machine and requests your chosen site. Snapshots and reports stay in your configured directory. Your AI client handles tool results under its own policies.

The crawler respects robots.txt by default. Optional JavaScript rendering requires Playwright and Chromium installed in your workspace.

Evidence needs context

Findings are checks and heuristics to investigate. A crawl cannot confirm Google indexing, rankings, traffic or complete site coverage. Review scan limits and use Search Console when you need indexing evidence.

Know what you install

Native ZIP and MCPB downloads are separate from npm. Read the client verification notes before installation. No marketplace approval or official platform endorsement is implied.

Put your next fix
to the test.

Start with a small crawl. Save the evidence. Check again after a change.