Skip to the content
Agent Tools

Support

Help with the Agent Tools ChatGPT plugins and how to reach us.

Install steps for ChatGPT, Claude, Cursor and Pi, each with a call checked against the live server, are on the install guides. Role-specific tasks are on the role guides.

Contact

Email support@openkrill.app.

Before you write, the status page shows whether each server passed its latest check and its latest real tool call.

When a tool starts to answer differently, the changelog lists every new tool and every change to what a tool takes or returns. How often our tools help an agent answer is measured on the accuracy test page.

When you write, say which plugin you used, what you asked and roughly when. Please do not send personal information.

Company Jobs Finder

Ask ChatGPT about open jobs at the companies you name, for example "what jobs is Stripe hiring for?", "is Notion hiring designers?" or "remote backend jobs at Ramp, Linear and Supabase with a posted salary". Up to five companies fit in one question. You can also describe your background briefly ("senior backend engineer, 7 years of Go, open to remote") and ask which jobs at up to three companies fit you best, or ask for a brief on one posting: its requirements, level signals, stated process and a tailoring checklist. You can also paste your own profile text (a headline and About section) and ask which words recruiters commonly search for a role are missing from it; the plugin never opens LinkedIn, so paste the text yourself. The profile is used only to answer that question and is not stored.

  • "No public job board found." The company does not publish its jobs on Greenhouse, Lever, Ashby, Workable or SmartRecruiters under its usual name. Try its careers page link instead of its name, or check its own careers page.
  • "This board has no open jobs right now." The board exists but lists nothing. Some companies keep an old board and post their jobs elsewhere.
  • A job that is gone or changed. Answers are cached for up to about two hours. The application link always goes to the employer's own page, which is the source of truth.
  • "Too many requests" or "daily call limit". The plugin is free to use and limits calls per network and per day. Wait a minute, or try again the next day (UTC).

Startup Name Check

Ask ChatGPT whether the domain for a name is taken, for example "is acme.com available?", "check if this startup name is taken" or "find an available .com for my bakery name". Up to five names or domains fit in one question.

  • "Available" is indicative. It is the status at that moment. It does not reserve the name, and a registration cost, when shown, is not a charge. Premium or reserved names can be taken or cost more.
  • "Unsupported". The ending is not covered by Cloudflare Registrar and has no public registration lookup (RDAP), so the plugin cannot say. Check a registrar.
  • "Unavailable". The lookup did not answer or limited our requests. Try again in a few minutes.
  • A name that was just registered or released. Answers are cached for up to five minutes.
  • Trademarks and owners. The plugin does not search trademarks and never shows who owns a domain. It shows registration status, dates when the registry publishes them, and a cost when Cloudflare Registrar returns one.

Package Health Check

Ask ChatGPT about an npm or Python package, for example "is lodash 4.17.15 vulnerable?", "is this npm package maintained?", "what license is this package?" or "safer alternative to request", or paste your package.json and ask it to check for known vulnerabilities. Send only public package names and versions, never source code or tokens.

  • "No npm package named ..." The name is misspelled, or the package is on the other ecosystem. Python packages are looked up on PyPI when you say so.
  • A version that "has no version ..." The registry has no such release. The answer names the latest version.
  • "Version ranges were checked at their lowest version." A package.json lists ranges such as ^4.17.15, not the installed version. The check uses the lowest version the range allows; a lockfile gives the exact answer.
  • Skipped dependencies. Dependencies that name a tag, url or workspace instead of a version are listed as skipped.
  • Release recency is not a verdict. A stable package can have few releases and still be fine, and a package with no known advisories can still be vulnerable.

Recall Check

Ask ChatGPT whether a vehicle or product has been recalled, for example "does my 2018 Honda Accord have any recalls?", "decode this VIN", "was there a recall on this infant formula?" or "dog food recall this week". It covers US data only and is not a safety guarantee.

  • "No recalls found". Vehicle recalls are looked up by year, make and model, not by VIN, so check nhtsa.gov/recalls with your VIN for the final word. A misspelled model is reported as not found rather than as having no recalls.
  • Car seats and tires. These are NHTSA recalls with no public lookup. Call NHTSA at 1-888-327-4236 or use nhtsa.gov/recalls.
  • Food and drugs. The FDA lists products by label wording, so try "infant formula" rather than "baby formula". Pet food comes from the FDA's 20 most recent announcements only.
  • "Service unavailable". A government source did not answer. Try again later.

Citation Checker

Paste a bibliography or a single reference into ChatGPT, for example "are these references real?", "check my bibliography", "verify this citation", "is this DOI correct?" or "has this paper been retracted?". Up to ten references fit in one question.

  • "Found". A real record matches the reference. A retraction status of retracted, expression of concern or corrected is shown with the notice dates when one is registered.
  • "Mismatch". A real paper exists but the DOI, title, authors or year in the reference differ. The record shown is the real one. A year-only mismatch can come from a preprint or a later edition.
  • "Not found". Nothing matching is in Crossref or OpenAlex. The reference may be invented or garbled, but books, reports, web pages and talks are often missing from these indexes, so check by hand.
  • "Unavailable". A lookup service could not be reached for that reference. Try again in a few minutes.
  • "Not checked" for retractions. The record comes from a source without retraction data, such as an arXiv or dataset DOI.
  • Slow answers. The scholarly indexes ask for about one request a second, so a full list of ten references can take half a minute.

Business Days & Workday Calc

Ask ChatGPT, for example, "how many working days between 14 July and 18 August in Germany?", "what date is 37 business days after 11 August in England?", "is Monday a bank holiday in Bavaria?" or "school holidays in Bavaria 2027".

  • How days are counted. Every answer lists its conventions: Saturday and Sunday are the weekend unless you name another, both the start and end dates of a range are counted unless you say otherwise, and when you add days the start date is day 0 and is not counted.
  • "Holidays only some regions keep". Without a region only nationwide holidays are applied. Name the state, province or nation (for example Bavaria or Scotland) to include its holidays.
  • "No holiday calendar for ...". The country code is not one of the two-letter codes the holiday source covers, or the year is outside 1976 to 2076.
  • No school holidays. School holidays are published for a limited set of countries, mostly in Europe.
  • A date that differs from your employer. The calendars are public lists. A company closure, a local observance or a contract rule is not in them, so confirm anything that matters, such as a legal deadline.

EU Business & VAT Checker

Ask ChatGPT, for example, "is the Austrian VAT number ATU33864707 valid?", "what is the LEI of Siemens Aktiengesellschaft?" or "look up Equinor in the Norwegian business register".

  • No name shown. Some countries, such as Germany and Spain, do not publish the name behind a VAT number. A name that could be a private person's (no legal form such as GmbH, SA, SRL or Limited) is withheld on purpose, and sole proprietors are left out of every result.
  • "Not valid". VIES does not know the number as an active VAT registration. Check the digits and the country; a number can be new, closed, or not registered for trade between EU countries. Great Britain left VIES in 2021, so only Northern Ireland (XI) can be checked.
  • "Unavailable". A member state's VAT service is sometimes down or busy. Try again in a few minutes.
  • Consultation number. VIES issues one only when you give your own VAT number with the check. It is proof of that check for tax authorities.
  • Valid is not trustworthy. It shows a registration, not that the business is solvent or the one you are dealing with. Compare the registered name with the invoice.
  • Registers. Company registers are France and Norway only; the LEI index covers companies of any country that have an LEI, which many small businesses do not.

US Federal Regulations (CFR)

Ask ChatGPT, for example, "what does 21 CFR 11.10 say?", "which regulations mention respirator fit testing?" or "what has the FDA proposed on food dyes?".

  • Not legal advice. It reads regulations and notices back to you; it does not tell you what applies to your situation. Ask a qualified lawyer before you rely on a result.
  • Section not found. Check the title number and the section number as written in the citation, for example 21 and 11.10. A section may be reserved or removed.
  • Cut text. A very long section is cut and says so; open the link to read the rest on ecfr.gov.
  • Not the official text. The eCFR is an unofficial edition updated daily. The legal text is the Federal Register and the annual CFR on govinfo.gov, and a Federal Register document may since have been changed or replaced.
  • Only US federal regulations. State and local law, statutes, court decisions and other countries' rules are not covered.

Site SEO & AI Crawler Audit

Ask ChatGPT, for example, "does example.com let GPTBot and ClaudeBot crawl it?", "which Open Graph tags is this page missing?" or "trace the redirects from http://example.com".

  • "Cannot be checked". Only public http and https sites on the standard ports are checked. Localhost, private and internal addresses, addresses with a login in them and other schemes such as file: or ftp: are refused, and so is a redirect that leads to one.
  • Not a security test. A robots.txt rule or an X-Robots-Tag header is a request to well-behaved crawlers, not a firewall. A clean result does not show that a site is protected, and a block does not stop a crawler that ignores it.
  • "Could not connect". The site was down, slow, or refused our request. Some sites block unknown clients, so a refusal is not proof that real visitors are blocked.
  • Missing tags. Only the HTML a server sends is read. Tags a page adds with JavaScript are not seen, and some social networks show their own preview regardless.
  • Old answer. Results are cached for five minutes, so a change you just made can take that long to show.
  • "ownership_not_proven". The exposure check returns no finding until you publish the token it gives: a DNS TXT record at _agent-tools-exposure. plus your host, or the file /.well-known/agent-tools-exposure.txt on the host. A bare IP address is refused, because the check cannot request one: point a host name at the server and check that name. Call again once the record or file is live; the refusal is never cached.
  • Ports not checked. The exposure check does not scan ports. The passive port sources do not allow use in a paid product, so open ports are listed under notChecked.

SEO Audit

Ask ChatGPT, for example, "audit example.com for SEO problems", "what SEO issues does https://example.com/pricing have?" or "is example.com ready to be named in AI search?".

  • "Cannot be checked". Only public http and https sites on the standard ports are audited. Localhost, private and internal addresses, addresses with a login in them and other schemes are refused, and so is a redirect that leads to one.
  • "robots.txt disallows". The audit follows the site's robots.txt for the user agent AgentToolsSEOAudit. If robots.txt blocks it, or cannot be read, no pages are fetched. Add an Allow rule for that user agent if you own the site.
  • A sample, not the whole site. At most 25 pages are read from the homepage, the sitemap and the links between them, so counts describe that sample. Pages added by JavaScript are not seen.
  • Heuristic rules. Rules marked as heuristics, such as title length, are conventions rather than search engine requirements. Passing every rule does not promise a ranking, and the tool does not measure rankings, traffic or backlinks.
  • Judgement skipped. The optional judgement (what a page is for, which pages compete) is off unless requested and is limited per day overall and per network. When it is unavailable or used up, you get the rule results and a note saying why.
  • Old answer. Results are cached for ten minutes, so a change you just made can take that long to show.
  • Local AI readiness. local_ai_readiness reads only your site and returns pass, warn or fail for service pages, location pages, visible prices, an FAQ, LocalBusiness markup, consistent contact details, on-site proof, a one-line description and recent dated content. It does not ask Gemini, ChatGPT or other engines who they name, and it does not open Google Business Profile, Yelp or other directories. Those are advice in the answer, not a lookup. It is limited per network and per day.
  • AI instructions. check_ai_instructions reads only the site you name and reports llms.txt, llms-full.txt, an AI or about-for-AI page, whether the home page links to them, and whether the file matches the site's name, products and facts. generate_ai_instructions returns a draft llms.txt for you to review. It does not publish the file. A public post is cited as the reason. Neither tool asks an AI engine who it names, and neither promises a change in AI Overviews.

Tariff & HS Code Duty Finder

Ask ChatGPT, for example, "what is the tariff code for a ceramic flower pot in the UK?", "what is the US duty on HTS 8507.60.00?" or "what duty applies to commodity code 6914100000 from Vietnam?".

  • Candidates, not a classification. The codes are matches from the official schedule. Which code is right depends on what the product is, what it is made of and how it is used, and customs decides. Confirm with customs or a licensed customs broker before you declare goods, or ask customs for a binding ruling.
  • "Nothing matched". The schedules match words in their descriptions. Describe the product in plainer words, such as the thing it is and the material it is made of, or try the noun alone.
  • A heading instead of a rate. A code that is only a heading has no duty of its own. The lines under it are listed so you can choose one and look that code up.
  • A rate that differs from your invoice. The duty shown is the rate in the schedule for that code. Quotas, anti-dumping and other trade remedies, excise, and for the US the additional duties in chapter 99 are not included, and rates change on published dates.
  • UK and US only. Other countries and the EU tariff are not covered.

Travel Check

Ask ChatGPT about official travel advice or an airport, for example "what is the travel advice for Thailand?", "what are the entry requirements for Japan?", "compare Mexico, Colombia and Peru" or "is JFK delayed right now?". Each answer shows the government's own text with its date and a link to the official page.

  • Whose advice it is. UK advice is written for British citizens and US advice for US citizens. Entry rules and risks can differ for other passports, and the plugin has no visa table, so check your own government.
  • One government only. Some places have advice from only one of the two governments. The answer says which one is missing.
  • "Fits more than one place". The name matched several countries, so the answer lists them. Ask again with one name.
  • Airports. FAA delays cover US airports only. Elsewhere only the weather is shown, and non-US airports are best asked for by their four-letter ICAO code, such as EGLL. It is not for flight planning.
  • "Unavailable". A government source did not answer. Try again later.

Food Facts

Ask ChatGPT about a packaged food, for example "what is in the product with barcode 3017620422003?", "find oat milks sold in France", "which of these barcodes has the least sugar?" or "does this product contain milk?". The data comes from Open Food Facts.

  • Check the label. Open Food Facts is edited by volunteers, so an entry can be incomplete, wrong or out of date, and recipes change. Every answer says how complete the entry looks. For an allergy, always read the physical label.
  • Allergen check. It reads only the allergens listed in the entry. "Unknown" means the entry lists no allergens and no ingredients, so nothing can be said either way.
  • "Not a valid barcode". The barcode must be 8 to 14 digits with a correct last digit. Check it for a typo.
  • "Not found". The product is not in the database. Coverage is best in France and Europe and thinner elsewhere.
  • "Unavailable". Open Food Facts did not answer. Try again later.

Car Check

Ask ChatGPT about a US car, for example "what car is VIN 1HGCM82633A004353?", "any recalls on a 2018 Toyota Camry?", "give me a report on a 2016 Honda Civic" or "what mpg does a 2022 Toyota RAV4 hybrid get?". The data comes from the US NHTSA and EPA.

  • By model, not by VIN. Recalls and complaints are listed for a year, make and model, so they cannot say whether your own car has an unrepaired recall. Enter the VIN at nhtsa.gov/recalls for that.
  • Complaints. A report shows only how many complaints owners filed and which components they name most. They are not confirmed defects.
  • "No ratings" or zero complaints. Many vehicles are not crash tested, and a different spelling of the model may list more. It is not proof that a car is safe.
  • Error codes. A VIN decode shows NHTSA's own error codes. Code 1 means the check digit is wrong, so look for a typo in the VIN.
  • "Not found". Try the base model name without a trim, such as "CR-V" instead of "CR-V EX-L". Car Check covers vehicles sold in the US only.
  • "Unavailable". An NHTSA or EPA service did not answer. Try again later.

Hire or Automate

Ask ChatGPT which parts of a job AI can do, which it should only draft, and which need a person, for example "which tasks of a bookkeeper can AI do?" or "write a job ad for the human part of a restaurant cook, plus a two-week trial". Send a job title, or an O*NET-SOC code. A job description is optional and is not stored. Do not send a person's name, email or phone number.

  • "No O*NET title matched." The title is not in the occupation tables. Try a more common title, or a code from onetonline.org.
  • Several occupations share the title. The answer expands the closest ones and lists the others. Add a short job description with no personal details, or pass the code you mean.
  • The human pile is empty. The rubric found no listed task that needs a person. That is a reason not to hire for those tasks, not a verdict on anyone.
  • "Too many requests" or "daily call limit". The plugin is free to use and limits calls per network and per day. The optional model tie-break has a separate daily cap. Wait, or try again the next day (UTC).

Buyer Signals

Ask ChatGPT, for example, "which software companies in California filed a Form D in the last two weeks?", "find Texas healthcare companies that raised money this month and have a bookkeeper role open 60+ days", "what do people complain about in monitoring tools?", "who is asking for an SEO audit on Hacker News or Stack Exchange?", "give me a weekly brief of bookkeeping asks" or "which sources does Buyer Signals use?".

  • "Daily limit". Searches are free and limited per day for each network and across all users. After the limit is reached (network_daily_limit or daily_limit), searches work again after 00:00 UTC. Listing the sources is never limited.
  • "Unknown region". Regions are any US state (name or two-letter code), the US, the UK, Canada, Israel, Ireland, Singapore, Australia and Hong Kong. Leave the region out to search everywhere.
  • No companies found. Only new Form D filings in the lookback window (1 to 60 days) are read, and funds and investment vehicles are left out. Try a longer lookback or a wider region.
  • No open roles shown. Roles are only checked when you name one, only on Greenhouse, Lever and Ashby boards found by company name, and only roles open 60 days or more are listed. Small and new companies often have no public board, and very large boards are skipped.
  • Funding is not intent. A Form D shows that a company raised money. It does not show that the company wants your product.
  • Missing sources. LinkedIn, Google, personal contact details and the Y Combinator directory are not used. The directory's terms prohibit scraping; ask for the sources list to see the terms links.

Send Out

Turn ChatGPT output into ready-to-send action links and files with no sign-in required, for example "draft an email to the project team with these action items", "make a calendar event for our sprint review on Monday", "create a WhatsApp link for this customer message", "export this note to my Obsidian vault", "prepare an X share link for this update" or "save this guide as a Markdown file".

  • Client-side review first. Every link opens your local application (email client, Google Calendar in browser, WhatsApp, Obsidian or X web composer). The draft is pre-filled so you can review and press send yourself.
  • Zero data custody. Send Out is completely keyless and stateless. No database records, no cloud storage and no message relays exist. Nothing is sent from our server and nothing is stored.
  • Link length warnings. If a mailto or web intent link is too long for certain browser or OS URL handlers (typically 2,000 characters), the tool notes it and provides a standard offline file (.eml, .ics or Markdown) instead.
  • Unsupported platforms. Send Out does not support Instagram direct messages, LinkedIn private messaging or automated email blasts.

For developers

The plugin is a public MCP server at https://jobs.openkrill.app/mcp (streamable HTTP, no key). Its tools, find_company_jobs, rank_jobs_for_profile, job_fit_brief and profile_keyword_check, only read public job boards. The profile passed to rank_jobs_for_profile and the text passed to profile_keyword_check are processed in memory and never stored or cached.

Startup Name Check is a public MCP server at https://names.openkrill.app/mcp (streamable HTTP, no key). Its single tool, check_domain_availability, only reads public domain registration data.

Package Health Check is a public MCP server at https://packages.openkrill.app/mcp (streamable HTTP, no key). Its tools, check_package and check_package_json, only read public package sources.

Recall Check is a public MCP server at https://recalls.openkrill.app/mcp (streamable HTTP, no key). Its five tools only read public recall data.

CertPacket is a public MCP server at https://certpacket.openkrill.app/mcp (streamable HTTP, no key, no account). Its tools, find_cert_rules, find_accepted_labs, check_lab_report, match_recalls, watch_recalls and stop_watching_recalls, read public CPSC, eCFR, Health Canada and EU Safety Gate data. Every answer is decision support only; confirm with your lab or customs broker before filing.

Photo Sheet is a public MCP server at https://photosheet.openkrill.app/mcp (streamable HTTP, no key, no account). Its tools, photo_to_sheet, voice_to_sheet and text_to_sheet, turn a photo, a voice note or a pasted text into a table with checked sums. A photo goes to Google Gemini and text to Groq or Gemini on their free tiers, so send nothing private. The table is kept for 24 hours; the photo and the audio are not kept.

SlotScout is a public MCP server at https://slotscout.openkrill.app/mcp (streamable HTTP, no key, no account). Its tools, scan_childcare_slots and compare_care_options, read the public Texas HHSC, New York OCFS and New York City child care licensing lists and Census tables, for one worksite in Texas or New York. Every answer is an educational estimate, not tax, legal or accounting advice; confirm the Section 45F credit with a tax adviser.

Citation Checker is a public MCP server at https://citations.openkrill.app/mcp (streamable HTTP, no key). Its single tool, check_references, only reads public scholarly indexes.

Live Card is a public MCP server at https://livecard.openkrill.app/mcp (streamable HTTP, no key). Its tools, render_live_card and live_card_status, turn self-contained HTML into a JPEG and MOV Live Photo pair plus an MP4 and a GIF. A file is available through a download link for 24 hours and is then deleted. If a link no longer works, ask the assistant to render it again. To have a file removed before it expires, write to the support address with the link. A status of external_url means the HTML pointed at a remote file. renderer_unavailable means the renderer is off. livecard_network_limit and livecard_global_limit mean the daily cap is spent until 00:00 UTC.

Document Tools is a public MCP server at https://docs.openkrill.app/mcp (streamable HTTP, no key). Its tools, count_text, make_calendar_file and make_csv_file, count the text you give them and write calendar (.ics) and CSV files. A file is available through a download link for 24 hours and is then deleted. If a link no longer works, ask the assistant to make the file again. To have a file removed before it expires, write to the support address with the link.

CVE Risk Check is a public MCP server at https://cve.openkrill.app/mcp (streamable HTTP, no key). Its tools, check_cve, check_cves and list_recent_kev, return a CVE's severity, CISA exploited-list status and EPSS score with the source and date of each, or the newest entries on the CISA exploited list. When the National Vulnerability Database asks this service to slow down, the answer says rate_limited and when to retry. If a record looks wrong, write to the support address with the CVE id and what you expected.

Domain Mail Check is a public MCP server at https://mailcheck.openkrill.app/mcp (streamable HTTP, no key). Its tools, check_email_auth, check_dns_records and lookup_domain_registration, return a domain's SPF, DMARC, DKIM and MX status with a grade, its DNS records, and its registrar and dates, with the source and date of each. A status of unavailable means a resolver or registry did not answer. If a result looks wrong, write to the support address with the domain and what you expected.

Taste Profile is a public MCP server at https://taste.openkrill.app/mcp (streamable HTTP, no key, no account). Its tools are add_refs, search_refs, describe_style, save_profile, get_profile and delete_profile. The first add_refs call returns a profile_key; keep it, because it is the only way back to the profile and cannot be recovered. A status of failed or partial lists each image with a code: image_unreadable means the link was not a public image, and daily_limit or ai_quota mean the free daily allowance is used up until 00:00 UTC. To remove your data call delete_profile with your key; to ask for help, write to the support address without the key.

Country Stats Facts is a public MCP server at https://countrystats.openkrill.app/mcp (streamable HTTP, no key). Its tools, get_country_indicator, compare_countries and get_imf_forecast, return a World Bank statistic for a country, a ranked comparison of up to six countries, and the IMF's projection for growth, inflation or debt, with the year, source and date of each. A status of no_data means the source has no figure; unavailable means it did not answer. If a result looks wrong, write to the support address with the country, indicator and what you expected.

Post Kit is a public MCP server at https://postkit.openkrill.app/mcp (streamable HTTP, no key). Its tools, draft_posts, draft_video_script, localize_post, launch_video and draft_tiktok_slides, write per-platform drafts checked against each platform's rules, write a 25-second short-form video script with the platform's video length limit and the AI-content label that applies when it declares an AI-generated person or voice, adapt your own post for another market, make a self-contained HTML launch clip from your own product page, and make 3 or 4 TikTok slideshows of 5 slides each as 1080x1920 PNG files with a caption and hashtags. A page at /slides on the same host does the same job. It never publishes. A status of capacity_reached means the shared free writing allowance is used up until 00:00 UTC; unavailable means the model did not answer; declined means the request was for a persona farm, a cloned voice or a fake-identity account; blocked_by_robots means the page's robots.txt disallows it. If a result looks wrong, write to the support address with the platform and what you expected.

Tax Numbers is a public MCP server at https://tax.openkrill.app/mcp (streamable HTTP, no key). Its tools, get_tax_parameters, get_mileage_rate, tax_deadlines and estimate_federal_tax, return US federal tax figures from a table checked against IRS documents, and a simple estimate. Each answer gives its sources and the date they were verified. If a figure looks wrong or out of date, write to the support address with the tax year, the question and the IRS document you checked.

Dev Facts Check is a public MCP server at https://devfacts.openkrill.app/mcp (streamable HTTP, no key). Its tools, get_eol_status, check_license, check_browser_support and lookup_rfc, return software end-of-life dates, SPDX license facts, browser support for web features and RFC records, each with its source and the date of the data. If an answer looks wrong, write to the support address with the product, license, feature or RFC and what you expected; we check it against the source, and a wrong date in endoflife.date itself is fixed there.

Paper Finder is a public MCP server at https://papers.openkrill.app/mcp (streamable HTTP, no key). Its tools, search_papers, search_arxiv and get_paper, return paper records from Europe PMC, Crossref and arXiv, each with its source and the date it was read. If an answer looks wrong, write to the support address with the query or identifier and what you expected; a wrong record is fixed at the index that holds it. To check whether a paper was retracted or whether a reference is real, use Citation Checker.

Market Indicators is a public MCP server at https://indicators.openkrill.app/mcp (streamable HTTP, no key). Its tools, compute_indicators and get_indicators, calculate indicators from candles you send or from the European Central Bank's daily euro reference rates, each with its source and the date of the last price. If a number looks wrong, write to the support address with the prices or pair and settings you used and the value you expected. It is informational only and not financial advice.

Store Price Tracker is a public MCP server at https://stores.openkrill.app/mcp (streamable HTTP, no key). Its tools, check_store_prices and get_product_price_history, only read the public product feed of Shopify stores and keep a daily record of that public product data. Store owners who want their store left out of the daily list, or who want the crawler to read less, can write to the support address with the store's address, or disallow StorePriceTracker in their robots.txt.

Business Days & Workday Calc is a public MCP server at https://businessdays.openkrill.app/mcp (streamable HTTP, no key). Its three tools, count_business_days, add_business_days and list_holidays, only read public holiday calendars.

EU Business & VAT Checker is a public MCP server at https://eubusiness.openkrill.app/mcp (streamable HTTP, no key). Its three tools, check_vat_number, find_company_lei and lookup_company_registry, only read public registers.

Tariff & HS Code Duty Finder is a public MCP server at https://tariffs.openkrill.app/mcp (streamable HTTP, no key). Its two tools, find_tariff_codes and get_duty_rate, only read the public UK and US tariff schedules.

US Federal Regulations (CFR) is a public MCP server at https://rules.openkrill.app/mcp (streamable HTTP, no key). Its three tools, get_cfr_section, search_cfr and search_federal_register, only read the public eCFR and Federal Register.

Site SEO & AI Crawler Audit is a public MCP server at https://sitecheck.openkrill.app/mcp (streamable HTTP, no key). Its first seven tools, check_ai_crawler_access, check_page_tags, trace_redirects, product_page_seo, ai_readiness_check, link_sample_check and landing_page_check, only fetch the public site you name, its robots.txt and, for the store checks, its llms.txt, agents.md, UCP file and sitemap, with up to 40 addresses from that sitemap. The landing page check reads one page and requests up to five of its link targets. These checks do not request a page that robots.txt disallows for Agent Tools. The eighth, exposure_check, reads public certificate transparency records from crt.sh and DNS answers from Cloudflare DNS over HTTPS, and requests only the proof file, the home page and 8 fixed paths of a host whose owner published its token. It returns no file contents and keeps nothing.

Page Fetch is a public MCP server at https://pagefetch.openkrill.app/mcp (streamable HTTP, no key, no account). Its tool, fetch_page, reads one public web page and returns its main content as Markdown with the title, description, language and links. It reads robots.txt first and skips a page that robots.txt disallows, and it never signs in, passes a paywall or solves a bot check. It keeps a page for 1 hour so a repeat call does not reach the site again. If a page comes back with menus or banners in it, or with text missing, write to the support address with the page address and what you expected.

Travel Check is a public MCP server at https://travel.openkrill.app/mcp (streamable HTTP, no key). Its tools, get_travel_advice, get_airport_status and compare_advisories, only read public government advisories and airport data.

Food Facts is a public MCP server at https://food.openkrill.app/mcp (streamable HTTP, no key). Its tools, lookup_barcode, search_products, compare_products and check_allergens, only read the public Open Food Facts database.

Car Check is a public MCP server at https://cars.openkrill.app/mcp (streamable HTTP, no key). Its tools, decode_vin, get_vehicle_report, list_recalls and get_fuel_economy, only read public NHTSA and EPA vehicle data.

Shop Duty Desk is a public MCP server at https://shopduty.openkrill.app/mcp (streamable HTTP, no key). Its tools, eu_duty_estimate, customs_data_gap_check, bulk_hs_candidates and landed_cost_estimate, only calculate from what is passed in, the UK and US tariff services and a small dated VAT table. Every tariff code it returns is a candidate, not a binding classification.

Industry Pulse is a public MCP server at https://pulse.openkrill.app/mcp (streamable HTTP, no key). Its tools, latest_events, daily_digest, search_events and product_feedback, answer what happened in AI coding agents (events grouped from publishers' feeds and Hacker News, with a score, a heat count and source links) and count the most common complaints and praises for a product. It reads RSS and Atom feeds, the Hacker News search API, GitHub issues and Apple App Store review feeds, and nothing that needs a login. An answer made right after a new deployment may say that the first collection is still running; ask again in a few minutes. The idea of scoring, grouping and digesting news comes from the open-source AIHOT project (MIT licence). Publishers who want a feed left out can write to the support address.

Catalog QA Desk is a public MCP server at https://catalogqa.openkrill.app/mcp (streamable HTTP, no key). Its tools, csv_import_preflight, merchant_feed_check, listing_text_audit and listing_fix_sheet, check a product CSV you pass in or, when you give a store address, read that store's public product feed once. Nothing is stored. Store owners who want the reader to skip their store can disallow CatalogQADesk in their robots.txt or write to the support address. Some large stores block the feed; upload the product CSV instead.

SEO Audit is a public MCP server at https://seoaudit.openkrill.app/mcp (streamable HTTP, no key). Its five tools, audit_site, audit_page, local_ai_readiness, check_ai_instructions and generate_ai_instructions, only fetch the public site you name, with its robots.txt respected. local_ai_readiness does not call an AI engine or a directory. generate_ai_instructions returns a draft and does not publish it.

AI Text Check is a public MCP server at https://aitext.openkrill.app/mcp (streamable HTTP, no key). Its tool, check_ai_writing, only matches fixed patterns in the text you pass in, up to 12,000 characters. It uses no AI model, does not rewrite anything and does not store the text. A finding is a place to look at, not proof of who wrote the text.

Tarot Reader is a public MCP server at https://tarot.openkrill.app/mcp (streamable HTTP, no key). Its tool, draw_cards, shuffles a 78-card deck and writes a short reflective note. It is entertainment, not a prediction. A crisis or medical question gets a pointer to real help instead of a reading. The question is not stored.

Dream Analyzer is a public MCP server at https://dream.openkrill.app/mcp (streamable HTTP, no key). Its tool, interpret_dream, offers common symbol meanings and reflective questions for a dream you describe. It does not diagnose, and it does not store the dream.

Feng Shui is a public MCP server at https://fengshui.openkrill.app/mcp (streamable HTTP, no key). Its tool, feng_shui_tips, suggests practical placement for a room you describe. It is not a promise of health, luck or money, and it does not store the description.

Dream Soulmate is a public MCP server at https://soulmate.openkrill.app/mcp (streamable HTTP, no key). Its tool, dream_soulmate, writes an imagined partner from answers you give. Words only: no photo, no face, and no match with a real person. The answers are not stored.

Daily Horoscope is a public MCP server at https://horoscope.openkrill.app/mcp (streamable HTTP, no key). Its tool, daily_horoscope, writes a short reflective horoscope note for a zodiac sign or birth date. It calculates signs and cusps from celestial positions, never predicts a certain future, and stores no birth dates.

Numerology Profile is a public MCP server at https://numerology.openkrill.app/mcp (streamable HTTP, no key). Its tool, numerology_profile, calculates core numerology numbers (life path, expression, soul urge, personality) with Pythagorean or Chaldean tables and master numbers. It is entertainment and reflection only, and stores no names or birth dates.

Birth Chart Snapshot is a public MCP server at https://birthchart.openkrill.app/mcp (streamable HTTP, no key). Its tool, birth_chart_snapshot, calculates sun, moon, and rising sign placements from astronomical algorithms. When birth time or place is omitted, it computes the sun sign only. It is entertainment and reflection only, and stores no birth dates or locations.

Sign Compatibility is a public MCP server at https://compatibility.openkrill.app/mcp (streamable HTTP, no key). Its tool, compatibility_reading, compares two zodiac signs or birth dates using astrological aspects, elements, and modalities. It is entertainment and reflection only, does not support relationship coercion or real person matching, and stores no birth dates.

Angel Numbers is a public MCP server at https://angel.openkrill.app/mcp (streamable HTTP, no key). Its tool, angel_number, looks up original meanings, root reductions, and mindfulness prompts for number sequences. It is entertainment and reflection only, refuses lottery and gambling queries, and stores no queries.

I Ching Oracle is a public MCP server at https://iching.openkrill.app/mcp (streamable HTTP, no key). Its tool, iching_cast, casts hexagrams and provides reflective ancient interpretations, judgments, and images from original texts. It is entertainment and reflection only, refuses wagering and relationship coercion, and stores no queries or casts.

Transcript Kit is a public MCP server at https://transcript.openkrill.app/mcp (streamable HTTP, no key). Its tools, transcribe_media, search_transcripts, suggest_clips and translate_subtitles, turn an audio or video file you upload, or a direct link to one, into a timestamped transcript with SRT and VTT captions, search your own transcripts of the last 24 hours, suggest 30 to 60 second clips as text, and translate captions into another language with a dub script. Files are limited to 15 MB and 15 minutes, and there are daily limits on minutes and on characters translated. Transcripts are deleted after 24 hours and the media is never kept. Only transcribe recordings you own or may use.

Video Grab is a public MCP server at https://videograb.openkrill.app/mcp (streamable HTTP, no key). Its tools, search_free_videos, own_video_export and video_metadata, return freely licensed video links with attribution, the official steps to export videos you uploaded, and public video metadata. A status of not_offered means the request was a platform download or a watermark removal, which this service does not do. Pexels and Pixabay stay off until their keys exist, and the answer says so. YouTube metadata uses the Data API when that key exists. If a licence looks wrong, write to the support address with the item page and what you expected.

LP Pool Check is a public MCP server at https://lpcheck.openkrill.app/mcp (streamable HTTP, no key). Its tools, check_pool, find_active_pools and estimate_lp, read public pool volume, TVL and fee APR from DefiLlama, and token flags from GoPlus when that chain is covered. A pool contract address is not in the index: if the answer says needs_tokens, pass the two token addresses or a pool id. Amounts you pass for an estimate are not stored. It is not financial advice and it cannot sign a transaction. If a figure looks wrong, write to the support address with the chain, the tokens or pool id, and what you expected.

Crypto Due Diligence is a public MCP server at https://cryptodd.openkrill.app/mcp (streamable HTTP, no key). screen_token reads DEX Screener, GoPlus, DefiLlama and GitHub and returns red flags, keep signals and the checks no free source can answer, such as the age of an X account. wallet_identity returns only names a wallet owner published: an ENS primary name that points back to the address and a Farcaster account that verified it. It never links a wallet to a private person. It is not financial advice and it cannot trade. To correct a name or a signal, write to the support address with the chain and the address.

Hire or Automate is a public MCP server at https://hire.openkrill.app/mcp (streamable HTTP, no key). Its tool, split_role, matches a job title to the O*NET 31.0 Database and tags each task. A job description is not stored. If a result looks wrong, write to the support address with the title and the occupation code you expected. It does not decide who to hire.

Ad Angle Map is a public MCP server at https://adangle.openkrill.app/mcp (streamable HTTP, no key). Its tools, map_store_offers, map_ad_angles and validate_angle_gaps, map offer and price angles from the public product feeds of Shopify stores you name, group ads you paste in, and check a phrase against Hacker News and Stack Exchange. An offer map is product pages and prices, not ads. Nothing is stored, and it never creates or runs ads.

Buyer Signals is a public MCP server at https://signals.openkrill.app/mcp (streamable HTTP, no key). find_buyer_signals and list_signal_sources read only public SEC filings and public job boards and return companies, never people. stack_pains counts tools, complaints, alternatives and switching triggers in public Hacker News and Stack Exchange posts, with a source link and a date, and never returns a username. find_buying_intent takes a service such as "SEO audit" and returns the posts that ask for it, each with a link, a date, the source, a quote under 15 words, an intent score, a label (intent or casual complaint) and a one-line reason; weekly_brief does the same for the last 7 days with a summary and counts. Scores come from plain rules, not from an AI model, and sellers are filtered out. Reddit is not read.

Trade Chart is a public MCP server at https://chart.openkrill.app/mcp (streamable HTTP, no key). Its tool, chart, draws a candlestick price chart from public prices and returns that chart as a page you can open. It does not place orders, connect a broker, or store the symbol you asked about. It is not financial advice. A chart page on the same host redraws the same public prices; it keeps no account of who opened it.

Send Out is a public MCP server at https://sendout.openkrill.app/mcp (streamable HTTP, no key). Its tools, draft_email, create_calendar_event, create_whatsapp_link, create_obsidian_note, create_x_post and create_markdown_file, generate ready-to-send intent links and offline files. It never sends messages from the server, never connects to external accounts and stores zero data. All drafts open in your local client for human review before sending.

Outreach Desk is a public MCP server at https://outreachdesk.openkrill.app/mcp (streamable HTTP, no key, no account). Its tools are set_criteria, score_message, get_scoreboard, draft_response and purge_data. Each ChatGPT user has one private desk, tied to the anonymized user id ChatGPT sends with every call, so there is no key to keep; a client that sends no user id cannot open a desk. Raw message text and recruiter contact details are never stored. Evaluation summaries are pruned after 30 days, and inactive desks are deleted after 90 days. To wipe individual senders or your entire desk, call purge_data.

Lane Planner is a public MCP server at https://lanes.openkrill.app/mcp (streamable HTTP, no key, no account). Its tools, plan_lanes and review_week, sort the interests you send into build, reading, open and parking-lot lanes, plan the next 3 to 7 days in minutes that add up to your hours, and turn a ticked checklist into the next plan. Nothing you send is stored or cached; the plan text carries everything the review needs. It is a planning tool only. If a plan looks wrong, write to the support address with the request you sent.

DESIGN.md Linter & Extractor is a public MCP server at https://designmd.openkrill.app/mcp (streamable HTTP, no key, no account). lint_design_md checks a DESIGN.md file with the official @google/design.md 0.4.0 linter and returns the same findings as npx @google/design.md lint. diff_design_md compares two versions token by token. extract_design_md writes a starter DESIGN.md from the CSS of a public page, after its robots.txt allows it. If a finding differs from the npx CLI, write to the support address with the file and the CLI output.

Etsy Listing Signals is a public MCP server at https://etsysignals.openkrill.app/mcp (streamable HTTP, no key, no account), with a page at /etsy-signals. Its single tool, find_etsy_breakouts, answers from readings our collector takes from the Etsy Open API every 4 hours: views and saves of new listings, their daily changes, a bucket and the top tags. Listings that name a trademark are counted and never shown. Until real readings exist it says not_started and shows no numbers. It never shows sales. If an answer looks wrong, write to the support address with the niche you asked for.

Email Engagement Report is a public MCP server at https://emailpld.openkrill.app/mcp (streamable HTTP, no key). Its tool, engagement_report, turns a pasted email event export into a per-recipient report. It does not store the export.

School Digest is a public MCP server at https://schooldigest.openkrill.app/mcp (streamable HTTP, no key), with a page at /schooldigest. Its tool, make_school_digest, builds one printable page per child from a snapshot you paste. It does not log in and does not store the snapshot.

Agent Trace Check is a public MCP server at https://agenttrace.openkrill.app/mcp (streamable HTTP, no key, no account). Its tool, analyze_trace, reads a tool-call log you paste from an AI agent run and returns the failed calls per tool, retry loops, repeated calls and slow calls. It reads only that text: it does not run the agent, call any tool in the log or fetch a URL, and it keeps nothing. If a call is paired with the wrong result or a format is not read, write to the support address with a short sample of the log and what you expected.

Prompt Injection Scanner is a public MCP server at https://injectionscan.openkrill.app/mcp (streamable HTTP, no key, no account), with a page at /injection-scan. Its tool, scan_text, checks text you paste, such as a web page, an email or a tool result, for instructions aimed at the model, hidden text and links that carry data out, and returns a verdict, a risk score and findings. It uses fixed rules, not a model, does not open the links it finds, and keeps nothing. A clean verdict means no known pattern, not proof of safety. If a line is flagged wrongly or an attack is missed, write to the support address with that short line and the verdict you expected.

IndexNow Submit is a public MCP server at https://indexnow.openkrill.app/mcp (streamable HTTP, no key, no account), with a page at /indexnow. indexnow_check reads a site's IndexNow key file, robots.txt and sitemap and submits nothing; without a key it makes one and names the file to publish. indexnow_submit sends the sitemap URLs to each IndexNow engine and returns every engine's answer, but only when the key file on the site holds exactly the key. One submit per site every 5 minutes; it keeps only the host name and the time of the last submit. Accepted means received, not indexed, and Google does not use IndexNow. If a key file is judged wrongly, write to the support address with the site and the verdict you expected.

Tenrung is a public MCP server at https://tenrung.openkrill.app/mcp (streamable HTTP, no key, no account), with a page at /tenrung. Its tool, score_ai_native_team, takes a team's own 0 to 3 score on 10 AI habits and returns the total out of 30, the band, the 3 first moves, a 90-day plan and the AI champion role card. It is a self-assessment, not an audit, and it keeps nothing: the result link holds only the 10 numbers and an optional team name. If a plan step reads wrong for a score, write to the support address with the result link and the step.

Namequota is a public MCP server at https://answershare.openkrill.app/mcp (streamable HTTP, no key, no account), with a page at /answer-share. Its tool, score_answer_share, scores AI answers you paste for one brand: whether each answer named it, linked its site or named a rival, with mention share and citation share across up to 50 answers. It calls no AI engine, opens no link and keeps nothing. If an answer is scored wrongly, write to the support address with the answer, the brand and the result you expected.

Inbox Document Sort is a public MCP server at https://docsort.openkrill.app/mcp (streamable HTTP, no key, no account), with a page at /docsort. Its tool, sort_inbox, classifies an email or PDF you paste as a purchase order, remittance advice, schedule change, vendor invoice, complaint or other, and returns sheet rows, a remittance check and a short briefing. It does not read a mailbox, run a paid OCR model, or store the file. A scan with no text layer needs the words passed as ocr_text. If a row looks wrong, write to the support address with the class you expected, not the whole file.

Remote Jobs by Country is a public MCP server at https://remotejobs.openkrill.app/mcp (streamable HTTP, no key, no account). Its tool, find_remote_jobs, searches a daily index of public company job boards and remote job feeds for jobs open to the country you name, with a verdict and a reason quoting each post. It stores only the public job posts, never anything about you. The same index is a page at tools.openkrill.app/remote-jobs. A verdict is a reading of the post, not a promise of being hired.

Trending Now is a public MCP server at https://trending.openkrill.app/mcp (streamable HTTP, no key, no account), with a page at /trending. Its tools, trending and trending_digest, answer from a store that reads Hacker News, DEV Community, GitHub, Product Hunt and Lobsters through their official APIs and public feeds every 2 hours: titles, links, counts, gain per hour, a heat score and rising keywords. Reddit and X are not read. If a list looks wrong, write to the support address with the niche and window you picked.

Belief Poll is a public MCP server at https://belief.openkrill.app/mcp (streamable HTTP, no key, no account). Its tool, poll_beliefs, asks Grok to search X and find Reddit threads on your question, reads every X post back through X's public embed endpoint and the Reddit comments from the thread's public feed, and returns the share holding each answer with a 95% range, the share on each source and the linked posts. A vote counts only when the post's own words that state the view are in the post, and one author counts once per source. It is a count of public posts, not a survey, and a poll is kept for 12 hours. If a post is sorted into the wrong answer, write to the support address with the question and the post link.

Post Stats is a public MCP server at https://poststats.openkrill.app/mcp (streamable HTTP, no key, no account). Its tools, get_post_stats, compare_posts and list_channel_stats, read the public numbers of YouTube and TikTok posts as a logged-out visitor would (X is refused, because its terms ban scraping and x.com/robots.txt disallows every path), and say where each number came from. They honour robots.txt, never sign in or pass a bot check, and keep an answer for 15 minutes. If a number looks wrong or a post comes back as shape_changed, write to the support address with the post address.