Local-first Python SEO evidence toolkit: analyzes Screaming Frog exports, adds bounded live and infrastructure checks, and produces structured audits, task backlogs, and reports via CLI and local MCP.
12 tools with mostly complete schemas and descriptions, but significant gaps in parameter documentation and error handling guidance. Tool names follow verb_noun convention well (seo_parse, seo_crawl_site, seo_domain_profile). Descriptions range 50-250 chars, meeting baseline. However, many parameters lack descriptions or constraints. Output schemas are not documented. Error recovery guidance is absent. No tool annotations visible (readOnlyHint/destructiveHint). Risk classifications (READ_ONLY vs WRITE) are present but not integrated into descriptions.
Describe the default crawl settings and their current values.
Crawl a site from a start URL by following links, or fetch an explicit urls list instead of following links at all, then audit the result through the same checks used for Screaming Frog exports. One of url or urls/urls_file is required. Same host only when following links, politeness adapts to the origin. Checks whose evidence a native crawl cannot produce are reported as skipped, never as clean.
Profile domain infrastructure: registration, DNS, hosting, ASN, country, and TLS.
Submit up to 10,000 changed URLs to IndexNow (Bing, Yandex, Naver, Seznam). Requires IndexNow key credential.
Parse SEO data (title, meta description, canonical, OG/Twitter, H1-H6, JSON-LD, links, visible text, word count) from one URL or a list of URLs.
Follow a live redirect chain for a URL and report each hop (status, location).
Output schemas not documented. Tools return complex nested structures (crawl results, audit artifacts, domain profiles) but LLMs cannot see what fields to expect. Forces agents to guess field names for downstream chaining.
Parameter descriptions missing or minimal. 'options' and 'overrides' params appear in multiple tools with only 'Additional parsing options' or 'Configuration overrides', no detail on what keys/values are valid. 'config' param in seo_crawl_site lacks any description.
No error recovery guidance. Tools like seo_crawl_site (WRITE, network-dependent) and seo_indexnow_submit (external API) lack descriptions of failure modes or what the agent should do if they fail. No mention of retryability, partial success, or fallback strategies.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | 2026-07-28+ | v2 |
Generate redirect rules (Apache mod_rewrite/Redirect, Nginx, or a custom template) from a list of {old_url, new_url} pairs.
Build a report from audit data in JSON, CSV, XLSX, Markdown, or DOCX format.
Check whether a URL is allowed by robots.txt.
Reparse retained HTML/DOM and rerun existing checks without network. Reads a native or derived SQLite scan and creates a new artifact with parent UUID and current analyzer provenance. The source is unchanged; an existing output is refused. Missing required bodies produce named unavailable errors. Missing live-only context remains skipped. No HTTP, DNS, browser, provider, or cache-miss fallback is permitted.
Compare two Screaming Frog exports or crawl artifacts to identify changes.
Import and audit a Screaming Frog XLSX export.
Tool annotations (readOnlyHint, destructiveHint, idempotentHint) not visible in schema. Risk field exists (READ_ONLY vs WRITE) but is not integrated into MCP tool definitions. Agents cannot infer which tools are safe to retry or have side effects.
Enum constraints missing. 'fmt' param in seo_redirects_generate and seo_report_build accept string values but no enum list is visible. LLMs will hallucinate invalid formats. 'robots' param in seo_crawl_site mentions 'respect, report_only, or ignore' in description but not as formal enum.