Squirrelscan

What squirrelscan does

squirrelscan provides website auditing tools for AI agents through a hosted MCP endpoint. Audits crawl a website and evaluate pages against more than 300 rules spanning SEO, crawlability, performance, security, accessibility, content, and related categories. Results include a health score, category scores, concrete findings, affected-page counts, recommendations, and report history.

The server also exposes an issue tracker for each website. Agents can list open issues, retrieve occurrence details and affected pages, and add Markdown comments for analysis or remediation notes. Separate entity tools inspect JSON-LD structured data as a site-wide graph, including declared entities, relationships, conflicting properties, missing identifiers, dangling references, and entity-related rule findings.

How it works

Call run_audit with a website URL to start an asynchronous cloud audit. The website is registered automatically on its first audit. Use quick, surface, or full to select the crawl level, or set max_pages for a smaller run; a one-page run audits only the entry URL. Poll get_audit_status with the returned run_id every 15–30 seconds until the run completes, then call get_report.

The default report format is structured JSON. It includes summary scores and top findings, while llm and markdown provide more extensive renderings. Reports may cap page and item samples, so agents should use the reported totals rather than list lengths. compare_audits identifies new, resolved, improved, worsened, unchanged, still-open, not-crawled, and not-checked findings between completed runs.

Recurring audits can activate after a website’s first successful audit on plans that support them. Scheduled runs consume credits. Audit estimates account for the base audit, crawled pages, and verified external links, and some settings changes require explicit confirmation before they take effect.

Setup and configuration

The repository includes skills and plugins for operating the CLI and running a website fix loop. The hosted MCP server can be installed through the Claude Code plugin, the Cursor plugin or MCP installer, OpenAI Codex configuration, or an Agent Plugins-compatible setup. Skills can also be installed with:

npx skills add squirrelscan/skills

The squirrel CLI must be installed and available on PATH for the repository’s skills. Verify it with squirrel --version. Authentication for CLI, CI, and MCP can use the SQUIRRELSCAN_API_KEY environment variable. create_api_key returns a new organization key only once, so it must be stored when created.

Tools and capabilities

The tool set supports:

• Starting, polling, listing, comparing, and reporting on audits.
• Registering, configuring, soft-deleting, and listing websites.
• Reading and updating crawl depth, scope, page limits, rendering, external-link checks, and recurring schedules.
• Listing rules and retrieving rule severity, recommendations, score weights, and documentation links.
• Browsing, inspecting, comparing, and exporting JSON-LD entity graphs.
• Tracking issues, retrieving occurrence details, and posting comments.
• Checking authentication, plans, balances, transactions, notifications, and audit costs.
• Sending feedback and creating organization API keys.

Limitations and notes

Only audit websites the user owns or is authorized to test. Audits spend credits as they run; a run may require confirmation when its estimate exceeds the organization’s automatic-run threshold or when a setting increases possible cost. A queued audit starts automatically, so it should not be submitted again.

A failed or blocked audit is not evidence that a site is healthy. Its report contains a failure reason and next step, and its health score may be null. A rule listed under notChecked must not be treated as passing. Full audits are restricted by plan, and plan limits can clamp requested page counts or levels.

The issue tracker is site-wide and deduplicated across runs: one failing rule across many pages can be one open issue. Its open count therefore should not be compared with a report’s issues_found value. For run-to-run comparisons, use the report health score or health_score returned for the run.