Is my domain ready for cold email? Our own score before and after one DNS record
We scored our own domain, openkrill.app, with Domain Mail Check before and after adding a DMARC record: the rules it tests, what moved the score, and what one record cannot fix.
What only this post answers: A real before and after on our own domain: 80 (almost) with no DMARC record, 96 (ready) the day we added one TXT record, and the rules that still matter.
Track a Shopify store's prices and stock from products.json: 3 real stores, measured daily
How to tell if a site runs Shopify, what its public products.json feed shows, and the day-over-day product, price and stock changes our Store Price Tracker measured on 3 real stores.
What only this post answers: Measured day-over-day changes on 3 real Shopify stores from their public products.json feed, plus a 1-request test for whether a site is Shopify.
How old are the open jobs on Greenhouse careers pages? We measured 5 companies
How to read open jobs straight from a company's Greenhouse or Lever board, and how many of 25 sampled postings at Stripe, Airbnb, Figma, Databricks and Anthropic are 60 days old or more.
What only this post answers: Measured posting ages from the public Greenhouse boards of 5 companies, and why an old posting is a question to ask, not proof of a ghost job.
The backend readiness checklist, explained with what an outside check can see
The 16 backend items an AI-built app often skips, why each one matters, the primary source behind it, and which 5 Site Check reads from outside versus the 12 that stay on your own list.
What only this post answers: Which of the 16 items an outside request can really see (5, one of them only in part) and which 12 stay on your own list.
We cut the text our MCP tools send back from 735,264 to 189,809 characters
What happened when we stopped echoing JSON in the text block of 113 MCP tools: the measured drop, which clients gain and which do not, and the one first-try fix that mattered most.
What only this post answers: Measured on 113 of our own tools: text fell 74%, and Claude Code users saw no change because it reads structuredContent.
We ran Site Check on our own tool site: one real finding, fixed the same day
Site Check read tools.openkrill.app from outside, found that plain http served the site, and turned clean after a one-rule fix. The run, the numbers and what it could not see.
What only this post answers: A real exposed-to-clean run on our own host: 34 requests per check, 1 medium finding, fixed with one redirect rule.
The tools these posts use are on the Agent Tools home page, and the install steps for each client are in the MCP install guides. Agents can read the same posts as text from llms.txt.