v0.3.0 — first public release

First public release of the OptimToken MCP server. The repo went public today under MIT; this tags what is running in production.

Connect it at https://ai-pricing-hub-mcp-9604f763.alpic.live/ — the server is hosted, there is nothing to install.

Five tools

Tool	What it answers
compare-llm-models	Browse and filter the catalogue on price, quality (Chatbot Arena ELO), efficiency and capabilities
recommend-llm-model	A ranked top 3 for one workload under your constraints, with a per-constraint satisfied/violated breakdown
compare-models-side-by-side	2 to 4 named models across all 8 use-case profiles at a chosen volume
estimate-llm-cost	Per-request and monthly cost for your own volume, token shape, cache hit rate and batch eligibility
compare-compute-pricing	Instance rates across AWS, Azure, GCP, OCI, OVH, DigitalOcean and Alibaba

All five are read-only, take no credentials, and store nothing.

Prices you can trust, or at least audit

• optimtoken.optimnow.io is the single source of truth. Its committed price archive, verified overrides table and anomaly check are what keep a half-price or double-price upstream row from silently halving your monthly figure.
• Three-tier fallback (live API, upstream direct, embedded snapshot). Tiers 2 and 3 serve uncorrected prices, so every response carries a provenance block with pricesVerified, and the lower tiers lead the answer with a notice.
• Costs are rounded at the MCP serialisation boundary, with tests and CI guarding the precision contract.

Client support

MCP registry manifest (io.github.OptimNow/ai-pricing-hub), explicit tool annotations for the Anthropic connectors directory, and MCP Apps widgets that render in ChatGPT: savings bars, a cost/ELO scatter, shared chrome and dark mode.

Caches are warmed at boot, so the first caller no longer pays for the rebuild.

Licence

MIT. Prices come from third-party sources and are provided as is: verify against vendor pricing pages before committing spend.