Usage Stats

Aggregate MCP tool usage. Public, because the honest number is the point.

This exists because "is anyone using this?" was unanswerable. Over one log window the MCP server served 139 successful POSTs and exactly ONE tool call: everything else connected, listed the tools, and disconnected -- directory health probers, not researchers. Connections are not usage, and only tool CALLS distinguish them.

Published rather than kept private: a project whose pitch is that it refuses to dress up what it does not know should not quietly sit on its own adoption numbers. Aggregate only -- no IP, no query text, no bill ids.

Our own read-path monitor calls a real tool every two minutes, on purpose: a handshake succeeds against a dead database, so nothing cheaper detects the outage that prompted it. That makes it ~720 synthetic calls a day, and for the first hours of this table's life it was 61 of 63 rows. Those calls are tagged at the MCP edge and excluded here, and the count we subtracted is reported as self_probe_calls so the arithmetic is checkable.