ClaudeSavvy is a free, open-source dashboard for Claude Code. It prices every token you burn, finds the cache savings you didn't know you had, and hands your agent harness an A–F report card.
Total spend, per-conversation, per-project, over time. Input, output, cache writes and reads — each priced to the cent.
See savings net of cache-write premiums — and get flagged when re-caching eats your budget instead of saving it.
Skills, sub-agents, plugins, hooks, commands — everything wired into Claude, inventoried at every scope level.
Each session graded A–F on how efficiently the harness ran — sorted worst-first so you know exactly what to tune.
Who's delegating to what. Track sub-agent fan-out and team usage so multi-agent runs stop being a black box.
Audit the 64 MCP servers you forgot you installed. See what's actually wired in, where, and what it's being used for.
ClaudeSavvy scores how efficiently your harness ran each session — then sorts the results worst-first. Tool errors, duplicate commands, repeated reads, scriptability misses: every point lost is a tuning opportunity found.
Runs locally against your own Claude Code data. No accounts, no telemetry, no bill — just charts.
