"Which model and which turns are eating the budget" is the preview, and it costs nothing. Pay only when you want the pattern detection and the recommendations on top. Worth knowing before you start: analysis reads your session logs, which contain your prompts — check what is in them before submitting real data, and use the sample report to see the shape first.
A spend analyser for AI agent session logs. It reads your Claude Code session files and reports what each model, tool and turn cost, so optimisation starts from measurement rather than a hunch.
- `agentcost_preview`, free: total cost broken down by model, token counts split into input, output and cache, tool usage frequency, and the top 5 most expensive turns
- `agentcost_sample`, free: a sample report, so you can see the output format before submitting real data
- `agentcost_analyze`, paid: waste-pattern detection, per-tool cost attribution and optimisation recommendations at $5, or the same plus industry benchmarks, savings projections and an exportable report at $15
The npm package alphaforensics-mcp, version 1.1.0, run as `npx -y alphaforensics-mcp`. No API key for the free preview tier. Your session logs live in `~/.claude/projects/`. Paid analysis is charged through x402 micropayments on Base mainnet — no subscription and no key to manage.
One command — npx -y alphaforensics-mcp