---
name: site-check-jobs
description: "Step-by-step job recipes for the Site SEO & AI Crawler Audit MCP server: Check product JSON-LD and page SEO; Check a landing page for conversion; Check AI readiness of a store or page. Use when the user asks things like \"Is https://www.allbirds.com/products/mens-tree-runners set up for search? Check the title, description and product data.\" or \"Review my landing page https://example.com/ for conversion. What should I fix first?\""
license: MIT
---

# Site SEO & AI Crawler Audit jobs

Each section below walks through one job with the tools of the `site-check` MCP server. The same recipes are MCP prompts (prompts/list) on `https://api.openkrill.app/gateway/mcp`.
Connect the server at `https://sitecheck.openkrill.app/mcp` in your MCP client settings (remote streamable HTTP, no key), then call its tools as normal tool calls. They are not command line programs: Bash, curl and scripts cannot reach them.

## Check product JSON-LD and page SEO

Job: Check product JSON-LD and page SEO, with the Site SEO & AI Crawler Audit MCP server.

Use the `product_page_seo` tool for this job.

Work on the user's request. If there is none yet, ask what they need. A request this job handles: "Is https://www.allbirds.com/products/mens-tree-runners set up for search? Check the title, description and product data."

Steps:
1. Call `product_page_seo` as a normal MCP tool call, with the request's values in the fields of its inputSchema. It needs `url`.
2. If a value the tool needs is missing and you cannot tell it from the request, ask the user one short question first.
3. This input is known to work; copy its format, not its values: `{"url":"https://www.allbirds.com/products/mens-tree-runners"}`
4. For the request "Is https://www.allbirds.com/products/mens-tree-runners set up for search? Check the title, description and product data.", a correct answer looks like this: Title and description lengths, h1 count, canonical link, the Product structured data (name, brand, SKU, GTIN, price, currency, availability, variants) and an issues list such as a missing GTIN or a canonical that points elsewhere.
5. Follow the server's own rules:
   - Pass a public page address or domain, such as https://example.com/pricing; a bare domain means https.
   - Private, local or internal addresses are refused: say so, ask for a public address and do not work around it.
   - Report results as what the site published at that moment. robots.txt is a request, not access control; if it is unreachable there is no verdict.
   - If robots.txt disallows a page for Agent Tools, tell the user and do not retry another way.
   - If the site did not answer, give the failure and suggest checking the address or trying later.
   - A link sample is not every page.
   - Content added by JavaScript is not seen.
   - Lead with exposed or clean and the top risks with one fix each, then the evidence, and say open ports are not checked.
   - Only confirmed findings have a severity.
   - For needs verification, give the unknown fact and how the owner checks it.
   - For legal signals, give law, finding, fix and source, list not checked ones with their reason, and say once it is not legal advice.
   - Pass one marketing email as email when the user asks about email.
   - Without proof it returns header results and the token to publish: give both.
   - Say database checks read only status and row counts, and that write rules are not tested.
   - For launch readiness, give launchList: the failing items with their fix first, then the warnings, and say the owner checks the rest; pass checklist true when the user wants that list.
   - For pre-launch pages, give shipNaked: each fail or warn with its finding, then its snippet as a code block to paste, and say once it is not legal advice.
   - Mention the offer line once at the end, never as a requirement.
   - Do not use these tools to scan for vulnerabilities, collect contacts, prices or content, or monitor a site.
6. If the call fails or the answer looks wrong, tell the user, then report it with `submit_feedback` (kind bug, tool `product_page_seo`).

## Check a landing page for conversion

Job: Check a landing page for conversion, with the Site SEO & AI Crawler Audit MCP server.

Use the `landing_page_check` tool for this job.

Work on the user's request. If there is none yet, ask what they need. A request this job handles: "Review my landing page https://example.com/ for conversion. What should I fix first?"

Steps:
1. Call `landing_page_check` as a normal MCP tool call, with the request's values in the fields of its inputSchema. It needs `url`.
2. If a value the tool needs is missing and you cannot tell it from the request, ask the user one short question first.
3. This input is known to work; copy its format, not its values: `{"url":"https://example.com/"}`
4. For the request "Review my landing page https://example.com/ for conversion. What should I fix first?", a correct answer looks like this: Twelve checks with pass, warn, fail or info, each with evidence and a fix when it is not a pass: the h1, the supporting sentence, the primary call to action and its repeats, link targets, form fields, trust signals, pricing or next step, viewport, HTML size, render-blocking hints and contact path, with a note that JavaScript-added content is not seen.
5. Follow the server's own rules:
   - Pass a public page address or domain, such as https://example.com/pricing; a bare domain means https.
   - Private, local or internal addresses are refused: say so, ask for a public address and do not work around it.
   - Report results as what the site published at that moment. robots.txt is a request, not access control; if it is unreachable there is no verdict.
   - If robots.txt disallows a page for Agent Tools, tell the user and do not retry another way.
   - If the site did not answer, give the failure and suggest checking the address or trying later.
   - A link sample is not every page.
   - Content added by JavaScript is not seen.
   - Lead with exposed or clean and the top risks with one fix each, then the evidence, and say open ports are not checked.
   - Only confirmed findings have a severity.
   - For needs verification, give the unknown fact and how the owner checks it.
   - For legal signals, give law, finding, fix and source, list not checked ones with their reason, and say once it is not legal advice.
   - Pass one marketing email as email when the user asks about email.
   - Without proof it returns header results and the token to publish: give both.
   - Say database checks read only status and row counts, and that write rules are not tested.
   - For launch readiness, give launchList: the failing items with their fix first, then the warnings, and say the owner checks the rest; pass checklist true when the user wants that list.
   - For pre-launch pages, give shipNaked: each fail or warn with its finding, then its snippet as a code block to paste, and say once it is not legal advice.
   - Mention the offer line once at the end, never as a requirement.
   - Do not use these tools to scan for vulnerabilities, collect contacts, prices or content, or monitor a site.
6. If the call fails or the answer looks wrong, tell the user, then report it with `submit_feedback` (kind bug, tool `landing_page_check`).

## Check AI readiness of a store or page

Job: Check AI readiness of a store or page, with the Site SEO & AI Crawler Audit MCP server.

Use the `ai_readiness_check` tool for this job.

Work on the user's request. If there is none yet, ask what they need. A request this job handles: "Will AI assistants cite this page, and what should I fix? Check https://www.allbirds.com/pages/our-materials-tree."

Steps:
1. Call `ai_readiness_check` as a normal MCP tool call, with the request's values in the fields of its inputSchema.
2. If a value the tool needs is missing and you cannot tell it from the request, ask the user one short question first.
3. This input is known to work; copy its format, not its values: `{"url":"https://maps.example.com/"}`
4. For the request "Will AI assistants cite this page, and what should I fix? Check https://www.allbirds.com/pages/our-materials-tree.", a correct answer looks like this: Evaluates five citation signals (summary clarity, author credentials, Q&A format, heading structure, structured data) with pass, warn or fail, evidence and a concrete fix, plus domain files and AI crawler rules.
5. Follow the server's own rules:
   - Pass a public page address or domain, such as https://example.com/pricing; a bare domain means https.
   - Private, local or internal addresses are refused: say so, ask for a public address and do not work around it.
   - Report results as what the site published at that moment. robots.txt is a request, not access control; if it is unreachable there is no verdict.
   - If robots.txt disallows a page for Agent Tools, tell the user and do not retry another way.
   - If the site did not answer, give the failure and suggest checking the address or trying later.
   - A link sample is not every page.
   - Content added by JavaScript is not seen.
   - Lead with exposed or clean and the top risks with one fix each, then the evidence, and say open ports are not checked.
   - Only confirmed findings have a severity.
   - For needs verification, give the unknown fact and how the owner checks it.
   - For legal signals, give law, finding, fix and source, list not checked ones with their reason, and say once it is not legal advice.
   - Pass one marketing email as email when the user asks about email.
   - Without proof it returns header results and the token to publish: give both.
   - Say database checks read only status and row counts, and that write rules are not tested.
   - For launch readiness, give launchList: the failing items with their fix first, then the warnings, and say the owner checks the rest; pass checklist true when the user wants that list.
   - For pre-launch pages, give shipNaked: each fail or warn with its finding, then its snippet as a code block to paste, and say once it is not legal advice.
   - Mention the offer line once at the end, never as a requirement.
   - Do not use these tools to scan for vulnerabilities, collect contacts, prices or content, or monitor a site.
6. If the call fails or the answer looks wrong, tell the user, then report it with `submit_feedback` (kind bug, tool `ai_readiness_check`).