AI ROI Calculator MCP Server

August 18, 2026 · View on GitHub

Built by OptimNow. Ask an AI assistant whether an AI project pays for itself, and get an answer built on live model prices, a 3-layer cost model, and arithmetic you can audit, instead of a plausible-sounding guess.

Connect in 30 seconds

The server is hosted, so there is nothing to install.

https://ai-roi-calculator-mc-e9dd36e7.alpic.live/mcp

Client	How to add it
	`claude mcp add --transport http ai-roi-calculator https://ai-roi-calculator-mc-e9dd36e7.alpic.live/mcp`
	Settings → Connectors → Add custom connector, paste the URL above
	Settings → Connectors → Add, paste the URL. The ROI dashboard renders as an interactive widget
	Add an HTTP MCP server entry pointing at the URL

Then just ask:

"We handle 10,000 support tickets a month. What's the ROI of deflecting them with Claude Haiku 4.5?"

Ask any general-purpose model to size an AI business case and it will happily produce a number. It will invent token prices, forget that retries are billed, ignore prompt caching and batch discounts, and skip the orchestration, retrieval and monitoring that make up most of the real bill. The answer looks confident, the arithmetic underneath is usually wrong, and there is no way to check it.

This server replaces the guess with a model you can inspect:

Layer 1, inference. Live per-model prices, including the provider's published prompt-cache read and batch rates. Retries are billed, because a retried call is charged.
Layer 2, harness. Orchestration, retrieval, tool APIs, logging, guardrails, egress and storage: the part everyone forgets, and often the larger half of the bill.
Layer 3, business value. 4 archetypes (cost displacement, revenue uplift, retention uplift, premium monetization), each with a realization rate, because not every technically-successful output turns into money.

Every figure it reports carries the price date it was computed from.

Tools

Tool	What it answers
`calculate-roi-v4`	"Is this worth doing?" Full ROI, payback, net benefit, cost breakdown, and a break-even volume for every value method except Retention Uplift — there value follows the customers you keep, not the volume, so no threshold exists. Renders an interactive dashboard in clients that support widgets.
`lookup-model-price`	"What does this model cost?" Live list, batch and prompt-cache prices for any model in the catalog, or the current top models.
`load-preset`	"What are sensible defaults for this use case?" Returns 1 of 11 preset scenarios without computing anything.
`sensitivity-analysis`	"What breaks this business case?" Impact ranking of volume, realization rate, cost and value at ±20%.

All 4 are read-only and take no credentials. Nothing you send is stored.

Presets: support, knowledgeQA, meetingSummary, marketingContent, codingTask, invoice, callSummary, agentWorkflow, recommendation, retention, premium.

Every answer ends with a deep link that opens the same scenario in the web calculator, so you can keep adjusting the assumptions in the browser and save the result.

Where the numbers come from

Prices are fetched live from the OptimToken catalog, which tracks 250+ models and refreshes daily from OpenRouter. If the hub is unreachable, the server falls back to an embedded snapshot and says so, keeping every figure attached to its provenance.

Formulas live in the AI ROI Calculator and are copied here verbatim by scripts/sync-engine.mjs. The web app and this server must answer the same question with the same number. They once drifted far enough that the same preset returned a 7-point different ROI depending on which one 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.

The full mathematical specification, including every formula and its rationale, is in METHODOLOGY.md.

Do not edit server/src/lib/: it is generated. Change the calculator, then run npm run sync:engine here and regenerate the goldens.

Local development

Requires Node.js 24+.

npm install
npm run dev              # Skybridge dev server + MCP inspector
npm test                 # engine goldens, catalog, formatting guards
npm run build            # widgets + server

Working on the calculation engine:

npm run sync:engine        # pull the current engine from the calculator
npm run sync:engine:check  # CI mode, fails if the local copy is stale
node scripts/generate-goldens.mjs

By default the sync looks for the calculator checked out at ../ai-roi-calculator. Override it with --from <path> or ROI_CALCULATOR_PATH.

ai-roi-calculator-mcp/
├─ server/src/index.ts        # tool + widget registrations
├─ server/src/catalog.ts      # live model prices, cached, snapshot fallback
├─ server/src/deeplink.ts     # links back into the web calculator
├─ server/src/lib/            # GENERATED, synced from the calculator
└─ web/src/widgets/calculate-roi-v4/

Built with Skybridge, deployed on Alpic.

The rest of the family


AI ROI Calculator	The web app. Same engine, full UI, saveable scenarios.
OptimToken	Compare what 250+ models cost per request, with caching and batch factored in.
cloud-finops-skills	FinOps knowledge for AI agents: AWS, Azure, GCP, AI inference, SaaS.
finops-mcp-resources	MCP servers, tutorials and client guides for cloud cost work.

License

Released under the MIT License.

The calculation engine in server/src/lib/ is generated from the AI ROI Calculator, which is MIT licensed as well. Prices come from third-party sources and are provided as is, without warranty. 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.

Questions about your own AI cost estimate? Talk to OptimNow.