Analyze Claude Code costs by model

Calculate detailed cost breakdown from Claude Code Agent Monitor

Weekly or monthly cost review when you want to understand spending by model and identify high-cost sessions for optimization.

• · Time period: 'today', 'this week', 'last 30 days'
• · Optional: session ID for per-session cost
• · Optional: budget constraint (e.g., 'budget $50/week')

• · Cost by model table (input, output, cache_read, cache_write tokens + cost)
• · Cost by session (top 10 most expensive)
• · Daily cost trend (7 or 30-day)
• · Token efficiency metrics (cache hit rate, compaction baseline recovery, output/input ratio)
• · Cost optimization recommendations

• · Agent Monitor API: /api/pricing, /api/pricing/cost, /api/pricing/cost/{sessionId}, /api/sessions, /api/analytics

• · Agent Monitor must be running (GET /api/health returns 200)
• · Pricing rules must be seeded (default: Opus/Sonnet/Haiku tiers)

• · Fails if Monitor API unreachable (must run 'npm start' in project dir)
• · Pricing rules can be customized but default tiers may not match actual rates
• · Session cost calculation depends on baselines — old sessions without baseline data show underestimated cost
• · Cache efficiency metric only meaningful if cache_write > 0

• · Four-endpoint API design (pricing rules, total cost, per-session cost, session list with inline cost)
• · Cost calculation formula explicitly shown: (input_tokens/1M) × input_per_mtok + ...
• · Longest model_pattern wins for specificity (e.g., claude-sonnet-4-5% matches exact version)
• · Baseline token preservation: old sessions' pre-compaction tokens tracked in baselines column
• · Default pricing tiers seeded for Opus/Sonnet/Haiku with input/output/cache-read/cache-write rates