GLLmolfar

A crawl starts from a domain or a section and finds URLs by following links, by reading the XML sitemaps, by both, or from a list you paste. Each response is recorded, and the findings are grouped by importance in a Fix list that says why each one matters, how to fix it and which URLs it covers. Batch fixes group findings that one change solves — site-wide, per template or per topic — so a template fix is one task rather than four hundred.

• Reachability and indexability: status codes, redirect chains, robots.txt, meta robots, X-Robots-Tag, canonicals, sitemap membership, click depth and orphan-page signals.
• On-page structure: titles with their pixel widths, meta descriptions, headings, duplicate values, and text that appears only after JavaScript rendering.
• Links and resources: internal and external links, anchors, broken URLs, links to redirects, images without alt text.
• Technical details: hreflang, structured data and JSON-LD validation, response and security headers.
• A single URL without a crawl: fetched, rendered, photographed on desktop and mobile, and checked — including accessibility checks with axe-core on the rendered page.

After the model is downloaded once, GLLmolfar turns page passages into embeddings locally and uses them for four things. The topic map lays the site out by meaning, with each group labelled by its distinctive terms. Cannibalization puts unusually similar same-language pages side by side with their shared passages and three ways to resolve them: merge and redirect, canonical, or make them different. Link opportunities pair a page that gets few internal links with related pages that do not link to it yet. Semantic search answers a typed question with the pages and passages closest to it.

The passage view shows how one page splits into the passages a retrieval system would store, and how closely each stays on the page's title and H1. All of these are leads for a person to review, not rankings: two pages close in meaning are worth a look, not proof that they compete in Google. With Search Console connected, the app also lists the queries for which Google actually showed two or more of your pages — that is what search really did.

A separate view asks whether AI crawlers can get in and read what they find: the robots.txt rules for each AI bot, whether the site has an llms.txt, whether it serves Markdown to agents that ask for it, and how well each page reads in passages an assistant could quote on their own. When GLLmolfar crawls as a bot preset such as GPTBot, it obeys that bot's own robots.txt group.

It reads the site. It does not measure whether assistants actually mention or cite you; that is a measurement over time, on questions you choose, and a different kind of tool.

Signed in, the app can run a local MCP server. Claude Code, Codex, Cursor or another MCP client can then start crawls, read the fix list and the URLs behind each fix, audit a page, search by meaning and save reports. Only programs on the same computer can reach it, with a token, and every step the agent takes shows in the app's window.

The app also ships 18 agent skills: ready jobs such as a technical audit, an AI-readiness review, a cannibalization clean-up, an internal linking plan or an llms.txt draft. Each skill tells the agent which GLLmolfar tools to call, in what order, and what to hand back. A fix can be turned into a Jira task from the app or by the agent.

Saved crawls of a site sit in one history, and any two can be compared — URLs, status, titles, headings, indexability, canonicals and meaning. Reports export as HTML or PDF for a client, a developer or a short summary. Crawls can be scheduled daily, weekly or monthly; they run on your computer while the app is open or in the tray, and the app says what got worse since the crawl before, including AI bots newly blocked in robots.txt. Any table can go to a new Google Sheet in your own Drive.

The crawl records and their history live in a local database in the app's data folder. Embeddings are computed on the device. A crawl still sends ordinary requests to the site being crawled, and optional checks of external links contact the linked sites. The app does not send the crawl to a cloud AI service.

What it does send to getlooploop.com is small and named on the product page: anonymous usage statistics (a random install ID, the app version, the operating system version, the language and how long the app stays open) and a check for a new version. Both can be switched off in Settings. Search Console data travels from Google to your computer.

GLLmolfar is in preview: it crawls real sites and finds real problems, and features, limits and file formats can change from one version to the next. The builds are not code-signed yet, so Windows and macOS warn before the first start. There is no Linux build and no build for Intel Macs.

It is downloaded from the product page through a personal link made for each download, never from a public file. A crawl has a page limit that is higher with a free account; the product page states the current figures, and they are an intention rather than a promise — they will change.

• It is not a rank tracker and does not forecast rankings; the site health score summarises weighted checks and is not a Google score.
• Similarity by meaning is a lead to review, not proof of keyword cannibalization.
• It does not measure whether AI assistants cite or mention a site.
• It does not change anything on the site it crawls.

The same term in the words somebody uses speaking to an assistant rather than typing into a box — written from the situation, which is why each one carries the situation it came from.

• GLLmolfar — product page
• Model Context Protocol
• axe-core — accessibility engine
• The /llms.txt file