Usage Board

① Headline figures

A "call" is one logical tool call from the server-side call log. "Independent callers" are deduplicated by salted hash — no raw address is involved.

② Last 30 days

Days are cut on server-local time; the vertical axis is rounded up.

Daily MCP tool calls

Daily record-page views (/record/*)

③ Most-called tools

Top 10 over the whole period, counting only the 59 published tools. A further 124 calls hit non-tool probe names (external scanners probing for tools that do not exist) and are excluded from coverage.

④ Visitor types

Classified by User-Agent over the whole request surface (incl. static assets). Our own availability monitoring and in-house test traffic fall under "Unclassified" — in-house requests total 23,972.

⑤ Most-cited records

A citation event means a caller used a specific record from a result (attributed by record ID). Each is verifiable at /record/<id>/.

⑥ Method and transparency

• MCP endpoint listed separately: /mcp and /mcp/… took 79,894 HTTP requests (protocol layer, incl. stream polling and handshakes); standalone discovery files (/mcp.json, /mcp/server-card) took a further 237. These and logical tool calls are different units and must not be added.
• Web surface: 293,536 requests in total (external 269,564 · in-house 23,972); page views 108,817 · static assets 53,612.
• Metering can be switched off: dropping an .off marker in the ledger directory stops recording without affecting the service.
• Reproducible: this page and /usage/index.json are generated from the same ledger by the same script.
• Cite this page: Trustbase Lab, Usage Board, as_of 2026-10-17. https://trustbaselab.com/en/usage/
• Anonymous vs keyed calls: 1,963 anonymous and 0 keyed calls in total (the two tiers offer identical functionality; a key changes no tool semantics). The auth field has been recorded since the access tiers shipped (2026-10-15); all earlier calls were anonymous by definition. Keys are issued via /en/keys/; no key details are shown on this page.

⑦ Classification rules & definitions

This section discloses how every classification and coverage figure on this page is computed. The rules apply at the aggregation/presentation layer only; the recording pipeline is unaffected.

• Definition of "independent sources": all deduplicated salted IP hashes seen on the web surface during the period — including crawlers, scanners and third-party monitors. It measures reach, not human visitors.
• Denominator of "tool coverage 59 / 59" = the currently published tool set (/mcp.json). 111 distinct tool names have appeared in the ledger overall; 52 of them are outside the published set: external negative probes (security scanners probing non-existent tools) 47, retired tool names get_benchmark, get_stats, query_fitness_studies, sandbox.execute_shell, protocol-level calls without a tool name 1. None of these count toward coverage.
• Improvement path: most unknown traffic is MCP client SDK default UAs (node, curl, aiohttp) and third-party monitors — UAs are self-reported and not enforceable. Once the access tiers (API key channel) ship, keyed calls will be attributed by key (auth field), which is far more reliable than UA; known monitor UAs will be added to the recognition list as they disclose themselves — but we will not loosen the classification standard to shrink the unknown share.