Librecrawl — Technical SEO Audit MCP Server

STRICT-MODE site audit — Screaming Frog parity. Same as librecrawl_audit but the response is annotated with an audit_complete flag and the run is marked incomplete if ANY of these happen: • crawl timed out before queue drained • max_pages cap was hit • upstream batch caps tripped • report file or sidecar CSV failed to write • sitemap reconciliation could not be computed Use this when you need a defensible audit certificate — every cap, sample, or partial result is surfaced loudly rather than hidden in summary counts. On success the run produces 3 files in REPORTS_DIR: • <domain>-…

Generate a full Markdown SEO audit report from a completed crawl. Runs 30+ checks (broken links, titles, metas, canonicals, images, orphan pages, redirect chains, analytics coverage, etc.) and saves as a .md file. USE THIS after librecrawl_start_crawl() finishes, or to re-generate a report from any past crawl_id via librecrawl_list_crawls(). Args: crawl_id: ID from librecrawl_start_crawl (optional — uses current crawl if omitted)

Get current crawler settings (maxUrls, maxDepth, crawlDelay, JS rendering, etc). Useful to confirm settings before starting a crawl. Returns the handful of settings that shape a crawl. Pass full=True for every setting, including the long default exclusion lists (about 2,500 tokens).

Poll current crawl progress. Repeat until is_running=False. Returns: is_running, crawled, queued, issues, base_url

Deep internal linking analysis — maps internal authority distribution across the site. Answers: - Which pages get the most internal links? (= Google considers them most important) - Which pages are orphans? (zero inbound internal links — invisible to Googlebot) - Which pages have zero outgoing internal links? (dead ends — no crawl flow out) - Which pages are crawl budget hubs? (link out to the most others) - What anchor text patterns dominate? (keyword signal analysis) USE THIS when asked: - "internal linking audit", "orphan pages", "link equity distribution" - "which pages need more interna…

List all saved crawls with URL, crawl_id, and timestamp.

NEW IN v1.5 — merge Google Search Console performance data into a crawl. Augments the .per-page.csv with 5 new columns (clicks_28d, impressions_28d, ctr_28d, position_28d, top_query) for every URL matched, and emits three new sidecar CSVs alongside the report: - <domain>-<ts>.gsc-winners.csv top 50 by clicks - <domain>-<ts>.gsc-losers.csv top 50 high-impressions + low CTR - <domain>-<ts>.gsc-quick-wins.csv position 11-20 + impressions > 100 GSC data shape (accepts what gsc-posi / aditya-gsc returns): { "rows": [ {"url": "https://...", "clicks": N, "impressions": …

Core Web Vitals and Lighthouse scores via Google PageSpeed Insights API. Returns LCP, CLS, FCP, TBT, INP, performance/SEO/accessibility scores, and real-user CrUX field data where available. USE THIS when asked: - "Core Web Vitals", "page speed", "Lighthouse score", "CWV check" - "how fast is [page]", "LCP / CLS issues", "performance audit" Args: url: Full URL to test strategy: "mobile" (default) or "desktop"

Batch Core Web Vitals audit — runs PageSpeed Insights on multiple pages, ranked worst to best performance score. Identifies your slowest pages first. Throttled to 1 req/sec to stay within the free 25k/day PSI quota. USE THIS when asked: - "check CWV across all pages", "which pages are slowest", "batch page speed audit" - "performance audit for the whole site" (pass top URLs from librecrawl_audit export) Args: urls: List of URLs to test (recommend top 10–25 pages) strategy: "mobile" (default) or "desktop"

Full PageSpeed Insights audit across EVERY crawled URL from a saved crawl. Makes the performance audit explicit — no sampling, no hidden caps. Each call to Google PSI counts against your daily 25k-call quota; budget accordingly (1 strategy = 1 call per URL). Use this when you need defensible CWV coverage for the entire site, not just the top N pages from librecrawl_pagespeed_audit. Args: crawl_id: The crawl ID to pull URLs from (librecrawl_list_crawls). strategy: "mobile" (default) or "desktop". limit: Hard ceiling on URLs to audit. 0 = no limit (recommended). …

Pause the currently running crawl. Resume with librecrawl_resume_crawl().

Return the contents of a saved audit report file. The MCP wrapper runs on a remote VPS, so callers cannot read REPORTS_DIR over the filesystem — this tool serves the file back through MCP. Path-restricted to REPORTS_DIR for safety; anything outside returns an error. Args: report_path: Path returned by librecrawl_audit / generate_report / full_audit_strict. Accepts the .md report, the .per-page.csv, or .sitemap-recon.csv. max_chars: Max characters to return. Default 200k. Content beyond this is truncated; set higher only if your client can handle it…

Resume a paused crawl in the current session.

Resume a previously interrupted crawl from the database — even from a NEW MCP session, days later, after server restarts. Continues from where the crawl left off, re-using all already-crawled URLs (no duplicate work, no extra load on the target). This is the way to handle "however big, in chunks" without overloading the target: Day 1: librecrawl_audit("https://huge-site.com", max_pages=5000) — runs until time/limits or you pause it Day 2: librecrawl_list_crawls() → find your crawl_id librecrawl_resume_from_crawl_id(crawl_id) — picks up where it left off, polite…

Schema.org / structured data coverage audit across multiple pages. No API key required. Identifies pages with no schema, shows schema type breakdown across the site, and flags missing rich result opportunities. NO SILENT CAPS: the whole list is processed. Internally batched in groups of `batch_size` with `batch_delay` seconds between requests to stay polite to the target server. Walks JSON-LD @graph containers (Yoast / RankMath / WPRM) so Recipe / Article / FAQPage types inside @graph are correctly extracted instead of being labelled "Unknown". USE THIS when asked: - "schema coverage across …

Extract and classify all Schema.org / JSON-LD structured data from a page. No API key required — fetches and parses the live page directly. Returns schema types found, which Google rich results they unlock (FAQ accordion, star ratings, breadcrumbs, product snippets, etc.), and which high-value schema types are missing. USE THIS when asked: - "does [page] have schema markup", "check structured data", "JSON-LD audit" - "what rich results is this page eligible for", "add FAQ schema" Args: url: Full URL to check

NEW IN v1.5 — validate JSON-LD structured data against schema.org spec AND Google Rich Results required fields. The existing schema_check / _audit tools EXTRACT and classify schema types. This tool goes further: for the 7 most common rich-result types (Article, Product, Recipe, FAQPage, BreadcrumbList, Event, JobPosting, VideoObject, HowTo) it checks that ALL required fields are present and surfaces specific missing-field findings. Output: writes <domain>-<ts>.schema-validation.csv alongside the report with one row per finding (url, type, severity, check, detail). USE THIS when asked: - "…

Instant technical SEO health check — no crawl needed, results in seconds. Checks robots.txt (rules, disallow, sitemap declaration, crawl-delay), sitemap.xml (existence, URL count, sitemap index), HTTPS redirect (http → https), and www/non-www canonicalisation. USE THIS when asked: - "check robots.txt", "is the sitemap set up", "quick site health" - "does [site] redirect http to https", "www vs non-www check" - Fast pre-audit sanity check before a full librecrawl_audit() Args: url: Site root URL (e.g. https://example.com)

NEW IN v1.4.0 — Start a chunked-progressive audit. Returns IMMEDIATELY with a session_id; the crawl runs in the background and survives PM2 restart. The runner polls upstream LibreCrawl every 20s, computes p95 latency + error rate per window, and re-tunes crawlDelay live (AIMD controller). Slow target → smaller per-second rate. Fast target → speeds up. Never overloads the site you're auditing. Returns: session_id — opaque id to poll librecrawl_audit_status() with status — initial state ("queued" → quickly becomes "crawling") url — echo of the input…

Start an async site crawl. Returns crawl_id immediately — does NOT wait for completion. Poll librecrawl_get_status() until done, then librecrawl_generate_report(crawl_id). Prefer librecrawl_audit() for a one-call full SEO audit with auto-report. Use this tool when you need step-by-step control or want to crawl in the background. Args: url: Full URL to crawl (e.g. https://example.com) max_pages: Max pages to crawl. Default 500. 0 = unlimited, which also needs confirm_unbounded=True. confirm_unbounded: Required with max_pages=0.

Stop the currently running crawl.

Get site link graph data from the current crawl — nodes (pages) and edges (links). Useful for understanding site architecture, identifying link clusters, and finding isolated sections of the site that Googlebot may not reach efficiently. Returns: nodes list (url, depth, status) and edges list (source → target links).

NEW IN v1.9.0 — Nuke ALL audit data: every session row, every artifact file in REPORTS_DIR, every LibreCrawl upstream crawl record. Use to return the MCP to a zero-memory baseline. REQUIRES confirm=True. Default refuses to fire. `success` is True only when every step succeeded; otherwise `wiped.errors` lists what is left.