OptimToken MCP
Built by OptimNow. Ask an AI assistant what a model or an instance actually costs, and get a dated, sourced figure instead of a number the model remembers from its training data.
Connect in 30 seconds
The server is hosted, so there is nothing to install.
https://optimtoken-mcp.optimnow.io/mcp
Paste it exactly as written, with /mcp and no trailing slash. The widgets
render only for that form: the host checks a hash of the URL you entered.
Then just ask:
"We send 200k support tickets a month at about 1,500 input tokens each. Which model gives me the best quality per euro, and what would it cost?"
Why this exists
Model prices change weekly, and a language model's idea of them is frozen at its training cutoff. Ask one what Claude or GPT costs and you get a confident answer that was true some months ago, with no date attached and no way to tell. The same applies to cloud instance rates, which additionally vary by region in ways nobody memorises.
This server replaces recall with a lookup:
- Live prices, not remembered ones. Every LLM figure comes from the OptimToken catalogue, which tracks 250+ models and refreshes daily.
- Corrected prices. OptimToken keeps a committed price archive, a verified overrides table checked against vendor pricing pages, and an anomaly check that alarms on the half-price and double-price breaks upstream feeds occasionally publish. This server asks that catalogue rather than re-deriving prices itself.
- Cost per request, not cost per million tokens. Price-per-token comparisons hide the thing you actually pay for. The tools apply your token shape, your cache hit rate and batch eligibility, and return a figure per request and per month.
- Every answer carries its provenance. Which tier served it, and whether the prices were verified.
Tools
All five are read-only and take no credentials. Nothing you send is stored.
Use case profiles ship with realistic token shapes, so you do not have to invent them: Support Ticket, Knowledge Q&A, Meeting Summary, Marketing Content, Coding Task, Invoice Processing, Call Summary, Agent Workflow.
Where the numbers come from
optimtoken.optimnow.io is the single source of truth. When it cannot be reached, the
server degrades in tiers rather than failing, and says which tier it used.
Tiers 2 and 3 serve uncorrected prices, and that matters more than it sounds. An
upstream feed once published a frontier model at half its real list price, which halves
every monthly figure derived from it. So every response carries a provenance object
with pricesVerified, and the lower tiers put a notice at the top of the answer. A
fallback should never quietly downgrade correctness.
Tier 1 is accepted only when the catalogue reports that it is itself serving fresh upstream data. If the site is on its own fallback, it carries no corrections, and this server treats it accordingly.
Local development
Requires Node.js 24+.
npm install
npm run dev # Skybridge dev server + MCP inspector at localhost:3000
npm test # schema conformance, serialisation precision, data sources
npm run build # widgets + server
The static fallback catalogue is refreshed by hand, not on a schedule:
npm run refresh-fallback
Because it is manual, check its dataAsOf before trusting a tier-3 response. An
unrefreshed fallback ages silently.
ai-pricing-hub-mcp/
├─ server/src/index.ts # tool + widget registrations
├─ server/src/lib/optimtoken-api.ts # the one base URL constant, fetch and timeout discipline
├─ server/src/lib/ # ranking, efficiency scoring, provenance, normalisation
├─ server/src/data/ # static fallback pricing + region maps
└─ web/src/widgets/ # React widgets rendered in the client
Built with Skybridge, deployed on Fly.io.
The rest of the family
License
Released under the MIT License.
Prices served by this server come from third-party sources and are provided as is, without warranty. Verify against vendor pricing pages before committing spend.
Questions about your own AI or cloud bill? Talk to OptimNow.