MCP context budget dashboard

Make AI coding agents' token spend and cost visible to individual developers

Problem

Heavy users of Claude Code and Codex spend $200–$400 a month on model APIs yet cannot say which task or context type burned the money. Someone even rebuilt the cost-estimation model for AI-driven development, implying line-count and COCOMO assumptions no longer hold. Bills arrive monthly, but costs cannot be attributed to projects or workflows.

Solution

Collect local agent session logs (token counts, models, tool calls), aggregate them by project, task type, and time window, and show daily/weekly cost curves with waste hints such as repeated huge contexts or inefficient tool loops. Everything is stored locally, with optional report exports for teams.

Users

Indie developers using coding agents daily, tech leads at small teams, freelancers delivering AI-assisted work

MVP

A local CLI or menu bar app that reads agent session logs; per-project token and cost rollups; daily/weekly views with budget alerts; flag unusually expensive sessions; CSV export

Revenue

Free for a single project, Pro one-time purchase or monthly plan for multi-project and team reports; enterprise per seat

Why now

Coding-agent subscriptions and API bills are now a developer's second-largest cost, yet independent cost-visibility tools barely exist and the logs already sit on local disk