HVTracker's trust layer is called by AI agents at decision time — over MCP and the public API. This page shows that traffic as it happens.
Free, read-only, no key. The same endpoint every call on this page came through.
claude mcp add --transport http hvtracker https://hvtracker.net/mcp
What's counted. The headline is tool calls: questions an
agent actually got an answer from, broken out per tool above. machine requests,
shown beneath it, are raw HTTP hits on /mcp, /api/v1/* and the
JSON data files. They measure traffic, not work done: about 98% are MCP protocol
handshakes and directory crawlers listing our tools, so they would overstate real use. Traffic from this page is excluded from both.
agent records are fetches of /data/agents/<slug>.json, the file
that carries each agent's signed trust credential; until October 2026 they were counted under other data files.
We record the tool name and the hour — never arguments, IPs, or anything identifying
who asked. Raw numbers: /api/v1/usage ·
API docs.