We measured the MCP registry for 40 days. The dataset is now something you can buy — and most of it you don't have to.
Free, forever, no signup (same as always):
- The full July-30 census — every one of the 10,716 advertised remote endpoints, one real JSON-RPC
initializeeach: census CSV - A 100-row sample showing the full column set
- Everything on the data page, corrections log included
The licensed package — $39 adds the parts that never got published:
- The daily reachability drift series: 400-endpoint samples drawn fresh from the live registry, every day since Aug 2. Population size at every draw. This is the only longitudinal reachability record of the registry we know of.
- A fresh re-probe of the full registry within 72 hours of purchase, so your copy describes the registry as it is, not as it was in July.
- Structured JSON of everything, the superseded-endpoint successor analysis (does churn have forwarding addresses? mostly no), a data card stating method and limits, and a single-buyer commercial license.
Buy the dataset — $39, instant download
Why trust it: we've published seven corrections to our own numbers, most of which made our findings less impressive. The corrections log ships inside the package. The probe_class column is the raw observation and bucket is our interpretation — kept separate so you can throw ours away.
Directory operators (PulseMCP, Glama, mcp.so, anyone with listings): we'll run your listings against the census and hand you your dead entries at no charge. Write to theopslog@proton.me.
Want a cut we didn't ship — by namespace, by transport, by failure mode? Same address. Measurement requests are usually a same-week turnaround.
Top comments (0)