AgentBill
See what every client's AI agents cost you, in dollars
https://agentbill.dev/mcpDetails
- Categories
- AIDeveloper ToolsAnalytics & Monitoring
- Target Audience
- AgenciesSaaS FoundersDevelopers
- Pricing
- Freemium from $29
Discovery signals
How AI and people discover AgentBill on PeerPush
About AgentBill
For freelancers and small agencies who build AI agents or automations for clients and pay the OpenAI or Anthropic bill themselves. AgentBill records what every model call costs in dollars at list price, split by client, agent and job. Once a workflow has a few days of history, you get an email the same day a client or agent spends 3x its usual day, instead of finding out when the invoice comes. You can also give any job a dollar ceiling: a call whose estimate would cross it gets approved: false before it runs, and your code or n8n workflow decides what happens next. Works with n8n, Make or any HTTP tool, Claude Code (two settings files, no code), Python and Node SDKs, and MCP clients. AgentBill never receives your prompts or your client's data, only a client label, the model name and token counts.
Product Video
Watch a video demo of AgentBill.
Screenshots
Reviews (1)
Average 5.0 out of 5
Based on 1 review
nice
Comments (3)
Getting the email the same day a client runs at 3x, instead of at invoice time, covers the case that actually hurts. And only sending a client label, model name and token counts makes it a lot easier to plug into client work.
Per-client baselines are smart. How do you handle negotiated provider pricing or cached-token discounts when real cost differs from list price? Custom rate cards would make agency margin reporting much more actionable.
Cached tokens already use the provider's cache rates. Negotiated discounts aren't modeled yet, so on a discounted account it reads as an upper bound. Rate cards are next: would a flat % per provider work, or do you need per-model rates?
Built this after a freelancer's workflow ran at 5x its normal cost for 21 days before he noticed. AgentBill tries to catch that on day one. Still early, blunt feedback welcome.