FREE COST CEILING CALCULATOR
Enter subagent count, tokens per agent, and model tier. See an estimated cost range per run, a monthly projection, how it stacks against Claude Pro/Max flat pricing, and a settings.json hook snippet to flag when you're over budget.
This free calculator estimates what a Claude Code run with multiple subagents actually costs: pick a model tier, set how many subagents fire and roughly how many tokens each one burns, and it projects a per-run cost range plus a monthly total at your expected runs/day. It compares that against Claude Pro and Max flat pricing so you know whether a flat plan or metered API billing fits your usage, and exports a Stop-hook snippet that estimates daily spend and flags a ceiling. Everything runs in your browser, with no signup and no upload.
Loading tool…
The actual input/output token split of a subagent run varies a lot by task — a research-heavy agent reads far more than it writes, while a code-generation agent does the opposite. The calculator computes the cost at a read-heavy split (85% input / 15% output) and a write-heavy split (60% input / 40% output) and shows both ends, since output tokens cost several times more than input tokens on every Claude model tier.
They use Anthropic's published per-model-tier API list prices (Haiku, Sonnet, Opus), each with a separate input and output rate per million tokens. Check anthropic.com/pricing for the exact current rate before budgeting against this number — list prices can change.
If a solo founder's subagent usage projects to, say, $40/month on metered API billing, a flat Claude Pro seat (~$20/mo) or Max plan (~$100 or ~$200/mo) can be cheaper and removes the surprise-bill risk entirely. The calculator flags which flat tier your projected monthly spend would fall under, so you can compare it against your actual plan instead of assuming API billing is the only option.
It's a Stop hook that runs after each turn, adds this run's estimated cost to a running daily total stored in `.claude/cost-log.json`, and exits with an error if that total passes the daily ceiling you set. It's an estimate-based tripwire, not a replacement for your real API usage dashboard — use it to catch a runaway multi-subagent loop before it burns a day's budget, not as an exact accounting tool.
No — enter the total number of agent invocations you expect in one run, including any nested subagents, since each one is a separate model call with its own token cost. If your workflow fans out dynamically, use your highest realistic count rather than the typical one.
No. All math runs in your browser. Nothing is uploaded or stored, and no signup is required.
By Rakshit Yadav · Published 2026-11-17 · Last updated 2026-11-17. Next: estimate the token cost of your CLAUDE.md, MCP servers, and skills, build the Stop hook this calculator exports, check which other rules still need enforcement, or see how much more a parallel agent team costs vs one agent.
[ RUNS IN YOUR BROWSER • DROP A FILE OR TAP GENERATE • NO SIGNUP ]
Go further
Engineering, Marketing, or Both: agents, skills, and slash commands you drop into any Cursor project.
Per run ≈ $0.38–$0.62
Monthly ≈ $115–$187 at 10 run(s)/day
Likely needs Claude Max 20x (~$200/mo flat) rather than API billing.
{
"hooks": {
"Stop": [
{
"hooks": [
{
"type": "command",
"command": "node -e \"const c=require('fs').existsSync('.claude/cost-log.json')?JSON.parse(require('fs').readFileSync('.claude/cost-log.json')):{spend:0,day:new Date().toDateString()};const today=new Date().toDateString();const s=c.day===today?c.spend:0;const next=s+0.6240;require('fs').mkdirSync('.claude',{recursive:true});require('fs').writeFileSync('.claude/cost-log.json',JSON.stringify({spend:next,day:today}));if(next>5){console.error('Daily cost ceiling of $5 exceeded (est. $'+next.toFixed(2)+'). Review before continuing.');process.exit(2)}\""
}
]
}
]
}
}Estimate only, based on list API pricing and an assumed input/output split — not a substitute for your real usage dashboard.