Agent user · start here

See your agent-host usage#

For the agent user who has not authored a connector. With zero setup — no defineConnector, no config file, no install — agent-connector usage scans each agent CLI's own native session logs/DBs and shows token usage aggregated by agent host, model, project, session, or day.

This track is three pages: Overview & quick start (this page — run it in one command), Reports & leaderboards (every usage flag, grouping, and export), and Coverage & confidence (which hosts are exact vs estimated, and the requires-sync rows).

This is the connector-free track

Everything here works straight from npx @ken-jo/agent-connector with no setup. usage reads your local agent-host logs read-only and never writes any host config, never runs install, and never needs a connector. Counts only — never your prompts or results.

Run it — zero setup#

No defineConnector, no config file, no install. If you simply want to know how many tokens your agent hosts are burning, run one command and agent-connector reads their own session logs read-only:

terminal
bash
# no install, no config, no defineConnector — just run it with npx:
npx @ken-jo/agent-connector usage report

# which agent host burned the most tokens? (ranks hosts)
npx @ken-jo/agent-connector usage leaderboard --by platform

# rank by model instead, scoped to the last 7 days:
npx @ken-jo/agent-connector usage leaderboard --by model --since 7d

# scope a report to a window / one platform; group however you like:
npx @ken-jo/agent-connector usage report --by day --since 7d
npx @ken-jo/agent-connector usage report --by model --platform claude-code

# export the raw aggregate rows (counts only — never prompts or results):
npx @ken-jo/agent-connector usage export --format csv --out usage.csv

What this can and cannot show

usage reports whole-conversation totals, not per-MCP / per-tool cost — see the canonical per-MCP vs host-scan explanation for why, and which track gives you per-tool numbers.