First public release of the AI ROI Calculator MCP server. The repo went public today under MIT; this tags what is running in production.
Connect it at https://ai-roi-calculator-mc-e9dd36e7.alpic.live/mcp — the server is hosted, there is nothing to install.
Ask an AI assistant whether an AI project pays for itself and get arithmetic you can audit, instead of a plausible-sounding guess built on remembered token prices.
Four tools
| Tool | What it answers |
|---|---|
calculate-roi-v4 |
"Is this worth doing?" Full ROI, payback, break-even volume, net benefit and cost breakdown, with an interactive dashboard in clients that support widgets |
lookup-model-price |
"What does this model cost?" Live list, batch and prompt-cache prices |
load-preset |
"What are sensible defaults?" 1 of 11 preset scenarios |
sensitivity-analysis |
"What breaks this business case?" Impact ranking of volume, realization rate, cost and value at ±20% |
All four are read-only, take no credentials, and store nothing.
The three-layer cost model
- Inference. Live per-model prices including published prompt-cache read and batch rates. Retries are billed, because a retried call is charged.
- Harness. Orchestration, retrieval, tool APIs, logging, guardrails, egress and storage: the part everyone forgets, and often the larger half.
- Business value. Four archetypes, each with a realization rate, because not every technically successful output turns into money.
Every figure carries the price date it was computed from.
One engine, two front ends
The calculation engine is generated from ai-roi-calculator and copied here verbatim. The web app and this server once drifted far enough that the same preset returned a 7-point different ROI depending on which you asked. CI now re-runs the sync against the calculator on every PR and weekly, and a golden-scenario suite fails if any preset's figures move.
Prices come live from OptimToken, with an embedded snapshot fallback that says so when it is used.
Licence
MIT, matching the calculator the engine is generated from. ROI figures are only as good as the assumptions you feed them: treat the output as a business case to challenge, not a forecast to bank on.