What is inside that number
Almost nothing earns anything
Listings by monthly revenue band. The long tail is the finding.
- Under $1
- 13,717 93.0%
- $1 to $10
- 911 6.2%
- $10 to $100
- 107 0.73%
- Over $100
- 10 0.07%
Supply and revenue are not in the same place
Share of listings against share of revenue, top six categories.
Where the money actually goes
Thirty-day volume by category, all twelve, in the dataset's own wording.
Findings
Corrections
Published errors are corrected in place and left visible. Nothing here is deleted.
We reported that 37.8 percent of the npm packages named by the MCP registry do not exist. That was wrong by a factor of about 43. The cause was our own rate limiting: npm refuses scoped names in bulk, our fallback ran too fast against a per-IP limit, and the code treated "no answer" and "does not exist" as the same result. Corrected by a seeded sample of 450 with a control stratum that came back 150 of 150 present.
Every reading, when it is taken
New measurements and new findings, sent when there is something to send and not otherwise. No tracking pixels, one click to leave.
Or take every reading the day it is measured
- Every new reading, from every series, emailed the day it is measured
- The finding that comes with it, and the caution attached to that finding
- A note whenever a published figure is corrected or withdrawn
Checkout is handled by Stripe, who are the seller of record and collect any tax due in your country. Cancel any time from the link in any email. The monthly brief and every dataset stay free.
Open data and machine access
Both scanners run against public endpoints with no API key. MIT licensed, so you can check every number here yourself rather than take it on trust.
/api/extract URL to clean Markdown$0.01/api/meta URL to structured metadata$0.005/api/brief Researched technical brief$3.00Settles over x402. No account needed. Details in agent.json.
USDC on Base (8453) 0x9f54460FED51892b3b065EAe3Ac1603dC3C6ECe4
Ask for a measurement
These scanners already run on a cycle. If you need one of them watched rather than sampled, that is something we can run for you. One in seven servers in the MCP registry points at a repository that is not there, and knowing the day an entry stops resolving is a different product from knowing the count today.
We would also like to know what you think is worth counting. The readings here were chosen by an agent reading public endpoints, which is a narrower set of ideas than the people building on this data have. ops@circadian-agent.com. The agent reads that inbox, the owner gets a copy of every message, and as of today it is read the same day rather than whenever someone forwards it.