Privacy policy

What our connectors keep, and what they do not.

Last updated October 11, 2026. Applies to Site Check by Modern Mustard Seed, Dry Run by Modern Mustard Seed, Idea to Product Sequencer by Modern Mustard Seed, Own It by Modern Mustard Seed, Package Pricing Lab by Modern Mustard Seed, Domain Health by Modern Mustard Seed, Local Trades Kit by Modern Mustard Seed, Small Business Funding Finder by Modern Mustard Seed, AI Search Kit by Modern Mustard Seed and Call Desk by Modern Mustard Seed at mcp.modernmustardseed.com, run by Modern Mustard Seed LLC, Kalispell, Montana.

What we receive

When Claude calls one of our tools, we receive only the tool arguments Claude sends: a website address, the sections or detail level asked for, a check id, and for the local presence score, the business name and any listing facts supplied (rating, review count, phone, address, hours, trade and source links). For Dry Run, the business figures Claude sends to describe a business (a preset name and numbers such as cash, costs, job values and payment terms, plus an optional business name). For the Idea to Product Sequencer, the idea description and the facts Claude sends about it (buyer, price, current alternative, evidence counts, feature lists, dates, and product metrics). For Own It, a domain or web address, an optional business name, and any profile links supplied. For Package Pricing Lab, the cost, capacity, deliverable and package figures Claude sends, and any package names and scope lines. For Domain Health, a domain name and any DKIM selector names supplied. For Local Trades Kit, a trade, a ZIP, town, state or climate region, a radius, a list of ZIPs or towns, a job type and a job completion date. For Small Business Funding Finder, the search words and filters Claude sends (applicant types, categories, agency, dates) and any opportunity id or number. For Call Desk, the phone numbers, states, call type and time Claude sends to be checked. We do not receive your conversation, your name, your email, your Claude account, or anything else from the chat.

For AI Search Kit, the arguments are the web addresses to audit or read, and the business facts you ask it to write into schema, llms.txt or compare across profiles (business name, phone, address, hours, links and similar).

There are no accounts, no sign-in and no cookies on the connector.

What the tools read

A scan reads only public information about the website you name: its homepage, robots.txt, sitemap.xml and llms.txt, its TLS certificate, its public DNS records (MX, SPF, DMARC), and the domain registry's public RDAP record. The tools refuse private and internal network addresses and never sign in to anything.

Own It reads only public information about the domain you name: the registry's and the registrar's public RDAP records, public DNS (nameservers, MX, SPF, A and CNAME records) through public resolvers, and the homepage as any browser receives it. Profile links you supply are classified by their address and never fetched.

Dry Run reads nothing outside our server. It runs arithmetic on the figures it is sent and returns the result.

The Idea to Product Sequencer reads nothing outside our server. It applies fixed rules to what it is sent and returns the result.

Package Pricing Lab reads nothing outside our server. It computes from the figures and lines it is sent and returns the result.

A Domain Health check reads only public information about the domain you name: its public DNS records (asked through the Cloudflare and Google public DNS resolvers), its TLS certificate, the headers of its homepage, its published MTA-STS policy and BIMI logo, and its registry's public RDAP record.

Local Trades Kit reads nothing outside our server. It answers from public Census, BLS and NOAA data and cited rules stored with the connector.

Small Business Funding Finder reads only the public Grants.gov API (api.grants.gov), sending it the search filters or the opportunity id. It never signs in and sends nothing about you.

AI Search Kit reads the public page you name plus that site's robots.txt, llms.txt and sitemap; to build an llms.txt it reads up to 30 public pages the site lists, honoring its robots.txt; for a consistency check it reads the public profile pages you list. Generating schema and validating pasted JSON-LD use no network at all.

Call Desk reads nothing outside our server. It looks numbers up in area code and rule data stored with the server; it never dials, texts or looks up a number with a carrier.

What we keep, and for how long

  • Scan results for a website are saved for 24 hours so a repeat question is answered instantly, then deleted automatically. A site that could not be reached is remembered for 10 minutes.
  • Domain Health results for a domain are saved for 5 minutes, then deleted automatically.
  • Usage counters (how many scans ran this hour and today, and when a site was last scanned) are kept only for the length of each limit window, at most one day.
  • One log line per tool call: the tool name, the domain scanned, how long it took, whether it succeeded, and whether the result came from the cache. Logs are kept for no more than 30 days.
  • Our hosting provider keeps standard request records (time, path, status and the calling network address, which for Claude is Anthropic's) under its own retention.
  • Business names and listing facts sent to the local presence score are used to compute the answer and are not stored.
  • Own It saves the facts it read about a domain (registration record, DNS answers, and the vendor names and public tag IDs detected on the homepage, never the page itself) for 24 hours, then deletes them automatically; an address that returned nothing is remembered for 10 minutes. Business names and profile links are used to compute the answer and are not stored. Its log line per call holds the tool name, the domain, how long it took, whether it succeeded and whether the result came from the cache.
  • Dry Run stores nothing about the business it simulates. Its log line per call holds only the tool name, the preset id, how many simulated years ran, how long it took and whether it succeeded; never the business name or any of its figures. Its usage counters are kept for the length of each limit window, at most one day.
  • Idea to Product Sequencer stores nothing about the idea, the features or the products it is sent. Its log line per call holds only the tool name, how long it took and whether it succeeded. Its usage counters are kept for the length of each limit window, at most one day.
  • Package Pricing Lab stores nothing it is sent: no figures, names or scope lines. Its log line per call holds only the tool name, how long it took and whether it succeeded. Its daily usage counter is kept for at most one day.
  • Local Trades Kit stores nothing it is sent. Its log line per call holds the tool name, a short label of the input (such as the trade and state, or the base ZIP and radius), how long it took and whether it succeeded. Its usage counters are kept for the length of each limit window, at most one day.
  • Small Business Funding Finder saves Grants.gov responses (searches for 10 minutes, listings for 60 minutes, the agency list for 6 hours) so repeat questions do not reach Grants.gov again, then deletes them automatically. Its log line per call holds only the tool name, how long it took, whether it succeeded and whether the result came from the cache; never the search words. Its usage counters are kept for the length of each limit window, at most one day.
  • AI Search Kit saves an audit for 10 minutes, an llms.txt draft built from a site for 30 minutes, and the name, address and phone read from a profile page for 30 minutes, then deletes them automatically. Facts you type in for schema, llms.txt or a consistency check are used to build the answer and are not stored.
  • Call Desk stores no phone number. Its log line per call holds only the tool name, the area code or state asked about, how long it took and whether it succeeded. Its per-minute counters are keyed the same way and its daily counter is a single total, kept at most one day.

How we use it

Only to answer the request, to enforce the usage limits, and to keep the service working. We do not use it for marketing, we do not build profiles, and we do not train models on it.

Who else touches it

  • Vercel hosts the connector and its logs.
  • Upstash stores the 24-hour caches and the usage counters for every connector.
  • The website you name, public DNS resolvers (Cloudflare and Google), and the domain registries and registrars receive the requests a scan, audit or check makes, as they would from any browser or mail server.
  • Grants.gov, run by the U.S. Department of Health and Human Services, receives the search filters and opportunity ids Funding Finder sends to its public API.

We do not sell or share data with anyone else.

Your choices

You can stop using the connector at any time by removing it in Claude. To ask what we hold about a website, or to have a saved scan deleted before its 24 hours are up, write to us and we will do it.

Contact

Modern Mustard Seed LLC, Kalispell, Montana. sarah@modernmustardseed.com. Security reports go to the same address.

If this policy changes, the date at the top changes with it.